Abstract: Since the constraints of various objective reasons and conditions, the lack of data exists in reliability engineering widely, especially in reliability-based design of aerospace. There is large amount subjective information exists in initialize data. Traditional reliability analysis method, such as probability-based method, can not handle the situation when both epistemic and aleatory uncertainties are presented in the system, the results acquired by probability-based method have large errors when used in practical engineering, sometimes may acquire a wrong result. This paper is aim at the spaceflight harmonic drive. After the sources of uncertainties and the method how to handle them are introduced, the mixture reliability analysis model of probability and non-probability are presented under both epistemic and aleatory uncertainties, and the reliability analysis model of harmonic drive is established while the sensitivity of parameters is given based on this reliability analysis model. Since the so called most probable point (MPP) search is not required, the proposed method is robust.