NIR and MIR spectroscopy applied to soil compositional analysis started to develop markedly in the 90's, taking advantage of various earlier advances in instrumentation and chemometrics for agricultural products. Today, NIR spectroscopy is envisioned as replacing laboratory analysis in certain applications such as soil carbon credit assessment at farm level. However, today's accuracy is still not ... [Show full abstract] satisfactory, compared with standard laboratory procedures, leading some authors to think that such a challenge will never be met. This paper investigates the critical points to be aware of when accuracy of NIR-based measurements is assessed. First, is the decomposition of the standard error of prediction (SEP) into a bias and a variance component, the latter being reducible by averaging, while the bias cannot. This decomposition is not used routinely in the soil science literature. Contrarily, a lognormal distribution of reference values is very often encountered with soil samples, such as elemental concentrations, e.g., carbon, with numerous small or zero values. These very skewed distributions make one take precautions when using inverse regression methods (such as PCR or PLS), which force the predictions towards the centre of the calibration set, leading to negative effects on the SEP - and therefore on prediction accuracy -, especially when lognormal distributions are encountered. Such distributions, which are very common for soil components, also make the RPD (Ratio to Performance Deviation) a useless and even a hazardous tool leading to erroneous conclusions. A new index based on the quartiles of the empirical distribution, RPIQ (Ratio of performance to inter-quartile), is proposed for overcoming this problem.