Pulse diagnosis with finger pulse-taking is popular in Chinese culture. Wrist pulse waveform analysis has becoming common in Traditional Chinese Medicine (TCM) engineering and diagnosis modernization. An improved two-step classification method is proposed in this paper to differentiate seven common TCM pulse conditions, include four mono and three concurrent pulses. For both time-domain and frequencydomain feature-based patterns, a total of ten effective discrimination functions (five for each domain ) are trained and tested for majority-rule based voting analysis. Case studies based on both basic one-step and improved two-step methods are given. Results show that the overall classification performance has been improved from 52% to 57% by introducing two-step method, and four out of seven individual classification accuracies are also improved.