We present an alternative method to analyse cosmic microwave background (CMB) maps. We base our analysis on the study of the partition function. This function is used to examine the CMB maps, making use of the different information embedded at different scales and moments. Using the partition function in a likelihood analysis in two dimensions (Qrms-PS, n), we find the best-fitting model to the best data available at present (the COBE–DMR 4 years data set). By means of this analysis we find a maximum in the likelihood function for n=1.8-0.65+0.35 and Qrms-PS = 10-2.5+3μ K (95 per cent confidence level) in agreement with the results of other similar analyses [Smoot et al. (1 yr), Bennet et al. (4 yr)]. Also making use of the partition function we perform a multifractal analysis and study the possible fractal nature of the CMB sky. We find that the measure used in the analysis is not a fractal. Finally, we use the partition function for testing the statistical distribution of the COBE—DMR data set. We conclude that no evidence of non-Gaussianity can be found by means of this method.Facultad de Ciencias Exactas