Chromatography Rf and Resolution Calculator

Chromatography: Rf for a TLC plate, resolution and plate count for a column

Two measurements students meet in the same term and rarely see side by side. For a thin-layer plate, how far the spot ran as a fraction of how far the solvent ran — the retardation factor RF. For a column, whether two peaks are actually separated: the resolution Rs from the retention times and the base widths, with the plate count for each peak and the rule of thumb for when a separation is good enough to measure. The two peaks below move with your numbers, so you can watch them merge.

Resolution, plate count, and the overlap you would actually get

Retention times and widths in, resolution out
In centimetres, measured from the pencil baseline to the centre of the spot, not its leading edge. If the spot is an elongated streak the centre is a judgement call and the RF is correspondingly rough — streaking usually means the spot was overloaded or the compound is interacting too strongly with the silica.
In centimetres, from the same baseline to the pencil line you drew along the solvent front the moment you took the plate out. Mark it immediately: the front is invisible once the plate dries, and a front guessed afterwards is the commonest source of a wrong RF. This must be larger than the spot distance, since nothing travels faster than the solvent carrying it.
In minutes, measured from injection to the apex of the earlier peak. Any consistent time unit works — seconds throughout gives the same resolution and the same plate count, because both are ratios. Retention volume works too, as long as the widths are volumes as well.
In the same unit, to the apex of the later peak. It must be larger than the first, because the resolution is defined with the later peak second; swap them and the answer comes out negative and the page refuses. The difference between the two is the whole numerator of the resolution.
The width at the baseline, in the same time unit: draw the tangents down the two sides of the peak and measure between where they cut the baseline. For a Gaussian peak that distance is 4σ. If your software gives you the width at half height instead, divide it by 0.5887 to get the base width — USP’s chromatography chapter is explicit that where there is any dispute, only the equations based on the base width are to be used.
The same measurement on the later peak. Peaks almost always get wider as they come off later, because a band spends longer diffusing on the way through, so a second width slightly larger than the first is normal and a second width smaller than the first is worth checking.
Your two peaks, drawn as Gaussians at the retention times and base widths you entered, with the apexes marked and the base widths measured off underneath. The shapes are to scale against each other, so as you narrow the widths or push the retention times apart you watch the pair separate — and as you widen them, watch the valley fill in and the two peaks become one. The time axis rescales itself to fit whatever you type.
1.54RsExample

A spot at 3.4 cm with the front at 7.6 cm; peaks at 4.20 and 4.80 min with base widths 0.36 and 0.42 min

Advertisement

Three IUPAC definitions, and the one convention that is not one

RF = distance of spot / distance of solvent front  ·  Rs = 2(tR2 − tR1)/(wb1 + wb2)  ·  N = 16(tR/wb)²  ·  overlap each side = Φ(−2Rs)
RF
the retardation factor: IUPAC’s “ratio of the distance travelled by the centre of the spot to the distance simultaneously travelled by the mobile phase”. Always less than one, conventionally quoted to two decimal places, and reproducible only for one plate, one solvent, one temperature and one chamber — which is why a literature RF is a hint and a co-spot with a known sample is proof
hRF
the same number times 100, which is how older and some European literature writes it. An hRF of 45 and an RF of 0.45 are the same measurement
Rs
peak resolution: the gap between the apexes divided by the average base width, which is where the factor of 2 comes from. Dimensionless, so any consistent unit of time or volume works. It is a property of the pair, not of either peak
wb = 4σ
the width at the baseline, between the tangents drawn down the sides of the peak. For a Gaussian that is four standard deviations, and the width at half height is 2√(2 ln 2)σ = 2.3548σ = 0.5887wb
N = 16(tR/wb)²
the plate number, a measure of column efficiency: how narrow a peak is relative to how long it took. IUPAC gives it three equivalent ways, (tR/σ)², 16(tR/wb)² and 5.545(tR/wh)², where 5.545 is 8 ln 2. All three assume a symmetrical peak, so a tailing peak’s plate count is a polite fiction
Φ(−2Rs)
the normal distribution function, which gives the fraction of one peak’s area lying past the midpoint between two equal Gaussians. This page evaluates it rather than quoting the familiar “99.7 per cent at 1.5”, so the number you see is for your own resolution

Worked example

A spot at 3.4 cm with the front at 7.6 cm; peaks at 4.20 and 4.80 min with base widths 0.36 and 0.42 min
THE TLC PART IS ONE DIVISION. RF = 3.4/7.6 = 0.447, or hRF = 44.7. That sits in the middle of the useful window, which is where you want it: a spot at 0.45 has spent real time on the silica, so a compound of slightly different polarity would land somewhere visibly different.
THE COLUMN PART: GAP FIRST. tR2 − tR1 = 4.80 − 4.20 = 0.60 min between the apexes.
THEN THE AVERAGE WIDTH. (0.36 + 0.42)/2 = 0.39 min. Resolution is the gap divided by that: 0.60/0.39 = 1.54. Written the usual way, Rs = 2(0.60)/(0.36 + 0.42) = 1.20/0.78 = 1.54.
SO WHAT DOES 1.54 BUY? Each peak spills Φ(−2 × 1.538) = Φ(−3.077) = 0.00105, so 0.105 per cent of each peak lies past the midpoint. Cut at the valley and each fraction is about 99.9 per cent of the right compound. That is baseline resolution with a little to spare.
PLATE COUNTS, ONE PER PEAK. N1 = 16(4.20/0.36)² = 16 × 136.11 = 2178 plates; N2 = 16(4.80/0.42)² = 16 × 130.61 = 2090 plates. They should agree, and roughly do; a large disagreement means one peak is not Gaussian, which makes both numbers suspect.
AND THE MARGIN. How wide could the peaks get before resolution fell to 1.5? 2(0.60)/1.5 = 0.800 min of total base width, against the 0.78 you have — so there is almost no room. A column that ages, a temperature that drifts up, or a sample injected in too much solvent would push this pair below baseline resolution, which is why a method is usually developed to about 2.0 rather than to 1.5.
Advertisement

What a resolution number actually looks like

RsOverlap each side, Φ(−2Rs)What you seeCan you integrate it?
0.421%one peak, slightly lopsidedno
0.78.1%a single peak with a shoulderno
1.02.3%two apexes, valley about half way downroughly, with large errors
1.250.62%a deep valley, tails still crossingyes, with a few per cent error
1.50.13%valley essentially at the baselineyes — the usual target
2.00.0032%a flat baseline between themyes, comfortably
2.50.000029%obviously separate peaksyes, and you have spent run time
The overlap column is computed, not quoted: it is the normal distribution function evaluated at −2Rs, which is the fraction of one equal-sized Gaussian peak lying beyond the midpoint between the two. The descriptions in the third column are the conventional readings of those numbers rather than measurements. The 1.5 target is a widely used convention; it is not stated as a universal requirement in the pharmacopoeial chapter cited below, where individual monographs set their own.

One idea behind both measurements

Chromatography of every kind works the same way. A mixture is carried by a mobile phase past a stationary phase, and each component spends a different fraction of its journey stuck to the stationary phase. Spend more time stuck and you travel more slowly. That is all. On a thin-layer plate the result is that different compounds end up at different heights; on a column it is that they come out at different times. RF and retention time are the same quantity seen in two geometries.

RF is simple to compute and easy to get wrong experimentally. It is the distance the spot’s centre moved divided by the distance the solvent front moved in the same run, so it is between 0 and 1 by construction. The trap is that it is only reproducible under conditions you probably did not control: the brand and thickness of the silica, the exact solvent composition, how saturated the chamber’s atmosphere was, the temperature, and how much material you loaded all move it. Two laboratories running what they both describe as 3:1 hexane / ethyl acetate can differ by 0.05 or more. This is why a literature RF is evidence and never proof, and why identification by TLC is done by running the unknown alongside a known sample on the same plate.

A column gives you more, because a peak has a width as well as a position. Resolution asks the only question that matters about a pair of peaks: is the gap between them large compared with how wide they are? Rs is the gap between the apexes divided by the average base width. Two peaks a long way apart can still be unresolved if they are broad, and two peaks nearly on top of each other can be resolved if they are sharp — which is what the diagram above is for. Watch what happens when you leave the retention times alone and halve both widths.

The plate count is the other half of the picture. It treats the column as though it were a stack of N equilibrium stages, and for a Gaussian peak it works out as 16(tR/wb)². The analogy is historical and the number is still the standard way to say how efficient a column is: a few thousand plates for a short HPLC column, tens of thousands for a long one, hundreds of thousands for a capillary GC column. Because both peaks in a pair travel the same column, their plate counts ought to agree; when they do not, one of the peaks is not Gaussian, and both plate counts — and the resolution computed from base widths — should be treated as approximate. The useful consequence of the plate model is that resolution grows as the square root of plate count, so doubling resolution by column length alone means four times the column, four times the run time and four times the pressure. Improving the chemistry is almost always cheaper.

The 1.5 rule deserves a word, because it is quoted as though it were a law. Take two equal Gaussian peaks and cut at the midpoint between them. The fraction of one peak that lands on the wrong side is the normal distribution function evaluated at −2Rs. At Rs = 1.5 that is Φ(−3) = 0.00135, so each of the two cut fractions carries about 0.135 per cent of its neighbour; the two spills added together come to 0.27 per cent, which is where the familiar 99.7 per cent and the phrase baseline resolution come from. It is a convention derived from an arithmetic fact, not a requirement handed down from a standards body. Pharmacopoeial methods set their own limits, often 2.0, and difficult pairs are sometimes accepted lower. The overlap row on this page does that calculation for whatever resolution you actually have, which is more useful than remembering one number.

If the point of separating the peaks is to measure how much of something is present, the peak area then has to be turned into a concentration, and that is a calibration problem rather than a chromatographic one: the calibration curve, LOD and LOQ page takes peak areas as readily as absorbances, and gives you the detection limit of the method as well.

Frequently asked questions

Why is my Rf different from the value in the paper?

Because RF depends on things the paper almost certainly did not specify: the silica’s brand and layer thickness, the exact solvent ratio, how saturated the developing chamber was, the temperature, and how much you loaded. Differences of 0.05 between laboratories are routine and do not mean either is wrong. Identify a compound by running it next to an authentic sample on the same plate, which removes every one of those variables at once.

Should I measure the peak width at the base or at half height?

At the base, if you have the choice, because that is the definition this page and the pharmacopoeial equations use — USP’s chromatography chapter says that in the event of dispute only the base-width equations are to be used. If your software only gives the width at half height, divide by 0.5887 to get the base width: for a Gaussian peak wh = 2.3548σ and wb = 4σ. The row above shows that conversion for your first peak.

My two plate counts disagree by 30 per cent. Which is right?

Probably neither, and that is the useful finding. Both peaks ran through the same column, so a real difference in efficiency between them is small; a large gap means at least one peak is not symmetrical, and every formula on this page assumes a Gaussian. Look at the peak shapes: tailing (a dragged trailing edge) usually means unwanted strong interactions with the stationary phase, and fronting often means the column is overloaded.

How do I actually improve a poor resolution?

Three levers, in the order worth trying. Change the selectivity — a different solvent strength, a different stationary phase, a shallower gradient — which moves the peaks apart and is almost always the cheapest win. Narrow the peaks: finer packing, a lower flow rate near the efficiency optimum, a smaller injection volume. Only then add column length, which buys resolution as the square root, so four times the column for twice the resolution.

What Rf should I aim for on a TLC plate?

Roughly 0.2 to 0.8, and 0.3 to 0.5 if you are choosing. Near the top of the plate everything polar bunches at the front and nothing separates; near the bottom everything stays at the baseline. The middle is where two compounds of similar polarity end up visibly apart. If you are picking a solvent for a column, a target of about 0.3 on TLC is the usual rule of thumb for the solvent to start the column with.

Does this page work in seconds, or in volumes?

Yes to both, as long as you are consistent. Resolution and plate count are both ratios of a retention measure to a width in the same unit, so seconds throughout, minutes throughout or millilitres of retention volume throughout all give identical numbers. The secondary rows are labelled ‘time units’ for that reason — they come back in whatever you put in.

Related calculators

References

  1. IUPAC, Compendium of Chemical Terminology (the Gold Book), entries “peak resolution” (P04465), “number of theoretical plates” (P04694) and “retardation factor” (R05353), all read 14 November 2026. Peak resolution is “the separation of two peaks in terms of their average peak width at base”, Rs = 2(tR2 − tR1)/(wb1 + wb2). The plate number is given three equivalent ways — N = (tR/σ)² = 16(tR/wb)² = 5.545(tR/wh)² — with the note that 5.545 is 8 ln 2, that all three assume a symmetrical Gaussian peak, and that the retention measure and the width must be in the same units. The retardation factor is the “ratio of the distance travelled by the centre of the spot to the distance simultaneously travelled by the mobile phase”, always less than one and usually quoted to two decimal places, with hRF = 100RF. The first and third are from the IUPAC chromatography recommendations, PAC, 1993, 65, 819, pages 845 and 847.
  2. United States Pharmacopeia, general chapter <621> Chromatography, the USP 32 / NF 27 text as reproduced on drugfuture.com (read 14 November 2026). Defines resolution R from the two retention times and the two base widths, and the plate number N for Gaussian peaks from the retention time and the base width — and states that “in the event of dispute, only equations based on peak width at baseline are to be used”, which is why this page asks for the width at the base rather than the width at half height. The equations themselves are images in that copy and could not be read as text, so they are quoted here from the IUPAC entries above rather than from USP. The threshold of 1.5 is not stated in this chapter as a universal requirement: individual monographs set their own resolution requirement. The 1.5 convention on this page is presented as what it is — a widely used rule of thumb, whose arithmetic the page derives rather than asserts.
  3. D. Harvey, Analytical Chemistry 2.1, section 5.4, “Linear Regression and Calibration Curves”, LibreTexts chem.libretexts.org (read 14 November 2026). The source of the regression arithmetic used here: the slope and intercept of an unweighted least-squares fit, the standard deviation about the regression sr = √[Σ(yi − ŷi)²/(n − 2)] — with the n − 2 explained as the two degrees of freedom spent on the slope and the intercept — and the back-calculation of an unknown as CA = (Ssamp − b0)/b1. Its worked example (six standards, slope 120.71, intercept 0.21, sr = 0.4035, unknown 0.241) was reproduced by hand while this page was being written, as a check that the four-standard shortcut used here agrees with the general formulae.
  4. COULD NOT BE READ, and nothing was substituted for it. The NCERT class XI and XII chemistry textbook PDFs on ncert.nic.in were refused by the server’s robots policy on 14 November 2026. That is why this page quotes no NCERT chapter, example or table number, and why the syllabus placement described in the prose is given in general terms rather than cited. If your own textbook’s worked example disagrees with this page in the third figure, check whether it is using a rounded atomic mass — 12 for carbon and 1 for hydrogen instead of 12.011 and 1.0080 — which is the usual cause and is not an error in either place.