The spherical Mexican hat wavelet is introduced in this paper, with the aim
of testing the Gaussianity of the cosmic microwave background temperature
fluctuations. Using the information given by the wavelet coefficients at se
veral scales, we have performed several statistical tests on the COBE-DMR m
aps to search for evidence of non-Gaussianity. Skewness, kurtosis, scale-sc
ale correlations (for two and three scales) and Kolmogorov-Smirnov tests in
dicate that the COBE-DMR data are consistent with a Gaussian distribution.
We have extended the analysis to compare temperature values provided by COB
E-DMR data with distributions (obtained from Gaussian simulations) at each
pixel and at each scale. The number of pixels with temperature values outsi
de the 95 and 99 per cent limits is consistent with that obtained from Gaus
sian simulations, at all scales. Moreover, the extrema values for COBE-DMR
data (maximum and minimum temperatures in the map) are also consistent with
those obtained from Gaussian simulations.