An intermittent fault diagnosis approach based on EMD-HMM to restrain false alarm for Builtin test (BIT) is proposed. Firstly, EMD is employed to decompose the original signal, the several energy dominating intrinsic mode functions (IMFs) are chosen and the energy feature parameters of each IMFs are extracted to form observed vectors of system. Then, the feature vectors are input into the trained HMM for malfunction recognition, the maximum log-likelihood probability is in the fault state. The results show that combining this feature vectors extracted method by EMD with HMM can diagnose intermittent fault effectively.