AbstractReflectance spectra of soil can be used to estimate the concentrations of organic carbon in soil (SOC). The estimates are more or less imprecise, but spectroscopy is quicker, less laborious and cheaper than conventional dry combustion analysis. Are the greater economy and efficiency sufficient to justify the loss of information arising from errors in estimation? We measured soil spectra with three instruments: a bench‐top mid‐infrared (mid‐IR) (mid‐IRb) spectrometer, a portable mid‐IR (mid‐IRp) spectrometer and a portable visible–near infrared (vis–NIRp) spectrometer. We calculated a quantity E to express the cost‐effectiveness of spectroscopic estimates relative to the conventional analysis, by accounting for their inaccuracy, their cost and their capacity, namely the maximum number of samples that can be prepared and measured daily. In all, 562 samples of soil were collected from 150 locations at four depths on a farm. The samples were dried and ground to particle sizes of ≤2 and ≤0.5 mm before measurements were made by dry‐combustion analysis. The machine learning algorithm Cubist was used to derive spectroscopic models of SOC concentrations and their uncertainties. We found that the mid‐IRb on the ≤0.5 mm samples was the most accurate and expensive but nevertheless sufficiently cost‐effective (large value of E) for determining the organic C. The mid‐IRp was somewhat more accurate, but its E was smaller than vis–NIRp on corresponding samples because it required more time to record the spectra. We also found that, with the portable spectrometers, the SOC predictions made on the ≤0.5 mm samples were somewhat more accurate than those made on the ≤2 mm samples, but their E was smaller because of the additional cost of sample preparation. The vis–NIRp on the ≤2 mm samples was the most cost‐effective for estimating SOC because it is cheap, accurate and has a large capacity for measurements.Highlights Concentrations of soil organic carbon (SOC) were determined by standard dry combustion and estimated from reflectance spectra recorded by three instruments. The labour required for each of the techniques and the cost, including that of the equipment, were recorded. A quantity E, expressing the cost‐effectiveness relative to dry combustion was calculated for each spectral technique, taking into account both accuracy and cost. Dry combustion was always more accurate than estimates from spectra for individual samples, and the technique was also more cost‐effective for small numbers of samples. The cost‐effectiveness of the spectral techniques varied among themselves, but all were more cost‐effective than dry combustion for large numbers of samples.