We present a method to detect non-Gaussianity in CMB temperature fluctuations maps, based on the spherical Mexican Hat wavelet. We have applied this method to artificially generated non-Gaussian maps using the Edgeworth expansion. Analysing the skewness and kurtosis of the wavelet coefficients in contrast to Gaussian simulations, the Mexican Hat is more efficient in detecting non-Gaussianity than the spherical Haar wavelet for all different leves of non-Gaussianity introduced. These results are relevant to test the Gaussian character of the CMB data. The method has also been applied to non-Gaussian maps generated by introducing an additional quadratic term in the gravitational potential.