We present a comparison between three approaches to test non-Gaussianity of cosmic microwave background data. The Minkowski functionals, the empirical process method and the skewness of wavelet coefficients are applied to maps generated from non-standard inflationary models and to Gaussian maps with point sources included. We discuss the different power of the pixel, harmonic and wavelet space methods on these simulated almost full-sky data (with Planck like noise). We also suggest a new procedure consisting of a combination of statistics in pixel, harmonic and wavelet space.