Near-infrared (NIR) reflection spectroscopy was used for in-line monitoring of the coating weight of printed layers of acrylate-based clear varnishes. Quantitative analysis of the spectral data was carried out by means of chemometric techniques based on partial least-squares (PLS) regression. The influence of variations of the chemical composition of the varnish formulations on the multivariate calibration models may lead to mispredictions of the coating weight. Therefore, it was compensated by special calibration procedures. Two different approaches were compared. Single-stage universal calibration included the different varnish formulations into one joint PLS model. Alternatively, the multistage procedure combined a preliminary identification step with subsequent quantitative analysis using specific calibration models for each varnish. The efficiency of both approaches was investigated by the prediction of the coating weights of independent test samples. Moreover, in-line measurements were carried out at a sheet-fed offset printing press. Both methods efficiently eliminated the influence of variations of the chemical composition. However, the multistage approach led to more precise predictions. In in-line monitoring trials during printing, the coating weight (in the range between 0.8 and 7 g m(-2)) was determined with a prediction error of 0.25 g m(-2) with the single-stage calibration model and 0.17 g m(-2) using the multistage approach.