We present a novel method for 3D shape representation learning using multi-scale wavelet decomposition. Distinct from previous works that either decompose 3D shapes into complementary components at a single scale, or naively adopt up-/down-sampling to build hierarchies and treat all points or local regions equally, we decompose 3D shapes into sub-bands components at multiple scales and all scales form a decomposition tree in a principled manner rooted in multi-resolution wavelet analysis. Specifically, we propose Adaptive Wavelet Transformer Network (AWT-Net) that firstly generates approximation or detail wavelet coefficients per point, classifying each point into high or low sub bands components, using lifting scheme at multiple scales recursively and hierarchically. Then, AWT-Net exploits Transformer that treats the features from different but complementary components as two integrated representations, and fuses them with the original shape features with different attentions. The wavelet coefficients can be learned without direct supervision on coefficients, and AWT-Net is fully differentiable and can be learned in an end-to-end fashion. Extensive experiments demonstrate that AWT-Net achieves competitive performance on 3D shape classification and segmentation benchmarks.