AbstractThe availability of good data is crucial to good reliability prediction. Two contrasting cases are noted: (1) when a large amount of experience with components is available but data are not collected and (2) when systems are so innovative that no reliability experience is available. The relationship of reliability to incompleteness of knowledge and the acausal nature of statistical/probabilistic models is discussed. The difficulty of assigning any absolute similarity between any one component and the members of a particular reliability set is outlined. It is concluded that engineering judgement must always play a vital role in reliability estimation.