A brain-computer interface(BCI) based on motor imagery (MI) translates the subject's motor intention into a control signal through classifying the electroencephalogram (EEG) patterns of different imagination tasks, e.g. hand and foot movements. Characteristic EEG spatial patterns make MI tasks substantially discriminable. Multi-channel EEGs are usually necessary for spatial pattern identification and therefore MI-based BCI is still in the stage of laboratory demonstration, to some extent, due to the need for constanly troublesome recording preparation. This paper presents a method for channel reduction in MI-based BCI. Common spatial pattern (CSP) method was employed to analyze spatial patterns of imagined hand and foot movements. Significant channels were selelcted by searching the maximunms of spatial pattern vectors in scalp mappings. A classification algorithm was developed by means of combining linear discriminat analysis towards even-related desynchronization (ERD) and readiness potential (RP). The classification accuracies with four optimal channels were 93.45% and 91.88% for two subjects.