Free graph maker & data converters

How to Make a Calibration Curve

By Plotiq Editorial Team · Published 14 November 2026 · Updated 14 November 2026 · 5 min read

A calibration curve turns an instrument signal, such as absorbance, into a concentration. This guide walks through the whole job with a seven-point example: plotting the standards, fitting the line, judging the fit and using it to read off an unknown.

What a calibration curve is for

Most instruments do not read concentration directly. A spectrophotometer reports absorbance, a chromatograph reports a peak area, an ion-selective electrode reports millivolts. To convert that signal into something meaningful you measure a set of standards, solutions whose concentration you know, and record the signal for each. Plotting signal against concentration and fitting a line gives you a rule you can apply to unknown samples.

The method only works when the relationship is predictable over the range you use. For absorbance, the Beer–Lambert law says the signal is proportional to concentration, so a straight line through the origin is expected at low to moderate concentrations. At high concentration real instruments bend away from the line, which is why you also need to know where the useful range ends.

Step 1: prepare and record the standards

Use at least five standards, and preferably six or seven, spread evenly across the range your unknowns are expected to fall in. Include a blank (zero concentration) so you can see the baseline. Prepare them independently if you can, rather than by serial dilution of one another, because a single pipetting error in a serial dilution carries into every later standard.

Record the concentration with its unit and the signal in two columns. The example below is a typical set of seven standards for a coloured compound measured at a fixed wavelength. The third column is the standard deviation of three replicate readings at each level.

Concentration (mg L⁻¹)AbsorbanceSD (n = 3)
00.0020.001
20.1010.004
40.2050.006
60.2970.007
80.4050.009
100.4980.011
120.6020.012

Step 2: plot it the right way round

Put concentration on the X axis and the instrument signal on the Y axis. Concentration is the quantity you control and know; the signal is the one that carries the measurement error. This matters because ordinary least squares assumes the X values are exact and minimises the vertical distances, so reversing the axes gives a slightly different line and the wrong error model.

Use a scatter plot with markers only, not a line connecting the points. A calibration is a fitted relationship, not a time series, and joining the dots hides how far each standard sits from the line. If you have replicate standard deviations, add them as error bars, as in the figure below.

0246810120.00.10.20.30.40.50.60.7Concentration (mg L-1)AbsorbanceAbsorbancey = 0.04989x + 0.002071, R2 = 0.9998
Seven standards with SD error bars and the least-squares line. Made with the Scatter Plot Maker.

Step 3: fit the line and read the numbers

A linear fit has the form y = mx + b, where m is the slope (sensitivity: signal per unit concentration) and b is the intercept (the signal at zero concentration). For the data above the least-squares result is a slope of 0.04989 absorbance units per mg L⁻¹, an intercept of 0.00207 and R² = 0.9998.

The intercept should be close to zero for a properly blanked method. Here it is 0.002, about the size of the blank reading, which is reassuring. A large positive or negative intercept points to a contaminated blank, a baseline offset or a standard that was prepared wrongly.

You can calculate the same numbers by hand to check any software: the slope is Sxy / Sxx, where Sxx is the sum of squared deviations of X from its mean and Sxy is the sum of the products of the X and Y deviations. Here Sxx = 112 and Sxy = 5.588, giving 5.588 / 112 = 0.04989.

Step 4: judge the fit, not just R²

R² near 1 is expected for a good calibration and many methods require 0.995 or better, but R² alone can hide problems. Always look at the residuals, the vertical distances between each point and the line. They should scatter randomly around zero. A run of positive residuals in the middle with negative ones at both ends is a smile-shaped pattern that tells you the data are curved and a straight line is the wrong model.

In this example the residuals are all within ±0.005 and change sign irregularly, so the line is appropriate. The standard deviation of the residuals (also called the standard error of the regression) is about 0.0034 absorbance units; you will need it for the detection limit below.

If the top standard falls below the line while the others sit on it, you have reached the end of the linear range. Drop that point, dilute your samples so they read inside the remaining range, and report the range you validated.

Step 5: calculate an unknown, the LOD and the LOQ

To find the concentration of an unknown, rearrange the equation: x = (y − b) / m. A sample that reads 0.350 absorbance has a concentration of (0.350 − 0.00207) / 0.04989 = 6.97 mg L⁻¹. Only report results that fall inside the range of your standards. Extrapolating beyond the highest standard assumes the line keeps going, which you have not checked.

The limit of detection (LOD) and limit of quantification (LOQ) come from the scatter around the line. A common convention is LOD = 3.3 × s / m and LOQ = 10 × s / m, where s is the residual standard deviation and m is the slope. With s = 0.0034 and m = 0.04989 this gives an LOD of about 0.22 mg L⁻¹ and an LOQ of about 0.67 mg L⁻¹. Quote the convention you used, because several are in circulation.

Presenting the calibration in a report

State the equation with its units, the R² value, the number of standards and the concentration range. Label the axes with quantity and unit, for example “Concentration (mg L⁻¹)” and “Absorbance (AU)”. If your instructor or journal wants a figure, export it as a vector file at the width of one column so the text stays crisp; the figure guide covers sizes and formats.

Frequently asked questions

How many standards do I need?
At least five, spread evenly across the working range, plus a blank. Six or seven is better. Fewer points make the fit sensitive to a single bad standard and give a poor estimate of the scatter.
Should the line be forced through zero?
Usually not. Forcing the intercept to zero removes a parameter that tells you whether the blank is behaving, and it biases the slope if the true intercept is not exactly zero. Fit the intercept and check that it is close to zero.
What if my calibration is curved?
Reduce the range to the linear region, or fit a second-order polynomial if your method allows it. Do not fit a curve just to improve R²; the model should match the chemistry or physics of the measurement.
Is R² of 0.99 good enough?
It depends on the method. Many analytical methods require 0.995 or 0.999. R² of 0.99 can still hide a clear curve, so check the residual plot as well.
Can I use the calibration on another day?
Only if the method has been validated for that. Instrument response drifts with lamp age, temperature and reagent batch, so most labs run fresh standards with every batch of samples.

Keep reading

Written and reviewed by the Plotiq Editorial Team. Spotted a mistake? Tell us and we will correct it. See About Plotiq for how these guides are made.