This study investigated the detailed kinematic characteristics, angular arc and peak angle of the pelvis, and the hip, knee and ankle joints of twenty-three spastic diplegic cerebral palsy (CP) children during level walking by three-dimensional analysis. All motion parameters were systematically and statistically analyzed with emphases on the significance of motion parameters at five specific gait events: heel strike, opposite toe-off, opposite heel strike, toe off and heel strike. According the differences in the kinematic pattern of the knee in sagittal plane, we categorized them into four CP groups: mild, crouch, recurvatum and jump. Through one-way ANOVA statistical analysis, temporal-spatial parameters and joint angles of the lower extremity were found to be significantly different among four CP groups. For the hip joint ankle, the crouch group had a tendency to keep flexed. And the maximum extension of crouch group was 14.8 degrees of flexion which was significantly lower (p<0.05) than the other groups. Also, the crouch had increased hip adduction and internal rotation throughout the gait cycle. For the ankle angle in the sagittal plane, the crouch group had a trend to dorsiflexion and the recurvatum group was to plantarflexion. The jump group was similar to normal group at loading response, but it later had a premature plantarflexion at initial stance. For foot progress angle, the jump group shown increased toe-in pattern. Ankle could interfere with the kinematic type of knee joint. If the ankle joint had excessive dorsiflexion during the stance phase of gait cycle, it would easily result in the crouch gait of the knee with crouch. But, if the ankle joint were equinus, it would easily result in the genu recurvatum or jump pattern of the knee. The contribution factors of the jump knee include maybe not only ankle equinus, but also harmstring spasticity. The jump group demonstrated greater hip flexion in initial stance, but later extends to 13.4 degrees. This is noteworthy different to Sutherland's finding.