Comparing with unimodal biometric recognition technology, multimodal biometric technology has better applicability, security and reliability, and has become the trend of biometric recognition. In the four fusion levels of multimodal biometrics, feature level fusion can extract more discriminative information, eliminate the redundancy among different features and reach the best recongnition effect in theory. However, the security of multimodal biometrics in feature level is rarely concerned. At present, the research of multimodal biometrics technology is mostly focused on bimodal biometric algorithms. This paper introduces the concept of intrusion tolerance and defines the measurement of intrusion tolerant capacity. The evaluation method of intrusion tolerant capacity is proposed to evaluate the tolerant capacity of the feature level bimodal biometric algorithms. Then the existing typical multimodal biometric algorithms are evaluated by the proposed method.