Exact diagnosis of electrical submersible pump （ESP）partial friction fault may avoid magnitude economic loss. Three-dimensional vibration accelerations of ESP were measured with acceleration transducer. Since working well ESPs are deeper in the earth, vibration signals seriously fade. This paper proposed that SVM is employed as classifier and wavelet parameters as features for ESP partial friction fault diagnosis. After SVM parameters selection with grid method, the highest recognition rate is up to 86.7%. Results indicate that SVM is competent not only for small sample recognitions but also for recognitions based on faintness signal. A new method is provided for ESP partial friction fault diagnosis.