The various methods for Traditional Chinese Medicine(TCM) pulse signal classification have extracted a large number of complex features, but it is difficult to use them efficiently in the classification algorithms due to the lack of systematic analysis on these features. This paper proposed a method for features evaluation and dimension reduction of pulse signal based on random forest. Firstly, the time domain, frequency domain and time-frequency domain features of pulse signal were extracted in 93 dimensions. Subsequently, the random forest algorithm was used to sort the importance of each feature based on the Gini index. Support vector machine (SVM)，back propagation neural network (BP-NN) and random forest (RF) algorithm were used to verify the correctness of the ranking. Finally, combined with the sequence forward selection algorithm, the feature selection was performed according to the classification accuracy of each algorithm. The experiments results showed that the ranking of the importance of these features based on the random forest algorithm was feasible, and after the feature selecting, the feature dimension decreased from 93 to 13. For the classification of normal, shi, wiry and slippery pulse, the accuracy of SVM and BP-NN increased by more than 10%, and the RF which is insensitive to feature redundancy also increased by 4.5%. As a result, this method can be used to a large number of features evaluation and dimension reduction in wrist pulse signal analysis, and improve the classification accuracy of the algorithm with efficiency.