- The stochastic component of a chemical sensor signal contains valuable information that can be visualized not only by spectral analysis but also by methods of high-order statistics (HOS). The analysis of HOS enables the extraction of nonconventional features and may lead to significant improvements in selectivity and sensitivity. We pay particular attention to the bispectrum that characterizes the non-Gaussian component and detects nonstationarity in analyzed noise. The results suggest that the bispectrum can be applied to gas recognition. The analysis of bispectra and the reproducibility statistics of skewness and kurtosis indicate that the measured time records were stationary.