Abstract: In order to improve the recognition effect of pulsar optical signals,an algorithm based on wavelet transform and bispectra analysis for pulsar signal recognition is proposed.First,the concept of bispectra in wavelet domain is proposed, namely,the bispectra of wavelet coefficients is calculated after signal is decomposed by wavelet transform.Then,the bispectral features of low frequency coefficients are extracted by PCA(principal component analysis),and those of high frequency coefficients are extracted by selected bispectra,according to the criterion of the maximal square sum of inter-cluster error. The bispectral features constitute feature vector.Finally,the minimum distance classifier is used to classify pulsar signals. The effectiveness of this algorithm in both pulsar signal recognition and feature dimensionality reduction is illustrated with some experimental results.