Astrophysical data sets are becoming more and more accurate making possible to address fundamental questions concerning how the Universe began and which processes governed its evolution. The answers to these questions point towards new physics which either takes place at very high energy scales, for instance inflation, or may require a revision of the Einstein theory of gravity. The impressive improvement of the quality of the data must be followed by a refinement of the tools we apply to analyze them, in order to be able not only to control all the systematics, but also to disentangle the primordial cosmological signal we are interested in from the emission of our Galaxy, like synchrotron and free-free radiation or thermal dust emission. We have been developing a tool, needlets, extremely suitable for the analysis of 2-dimension signal on the sphere. Usually such studies are carried out either by looking at the actual signal on the sky in a given direction or by taking its spherical harmonic transform. Most the times both approaches are combined to optimize the result. This procedure con be difficult and not very efficient if only a portion of the sky is observed. To solve this problem a wavelet-based approach has been proposed, which combines a multi-scale study together with a local analysis of field. Needlets are a peculiar type of wavelets whose scaling and localization properties are analytical determined and very sharp. This translates into very small correlation between the functions which is crucial for high level statistical analysis. We applied needlets to the WMAP CMB temperature maps to test the Gaussianity assumption of cosmological perturbations and the investigate dark energy properties.