When we consider the reliability of structural member group, we sometimes face the case of not being able to treat members as samples from the same population. For this case, reliability analysis becomes a very difficult one. In order to perform the analysis, we require subjective information which can describe the structure of member group and cover the shortage of objective probability data. The present paper deals with the problem of how to utilize effectively the subjective information in the structural reliability analysis. A method of utilizing pattern perception of human is presented. Expert's intuitions perceived from the insufficient objective data as well as his knowledge and experience are expressed by the “subjective distribution patterns” and utilized as the prior probability information in the reliability analysis. First, several patterns which can be the cause of objective data are generated by the expert's judgement. Second, in order to distinguish between probable distribution patterns and misdirected patterns, probabilistic analysis using likelihood function is carried out. The improvement of estimated pattern is also performed with the aid of Monte Carlo simulation. The effectiveness of this method is numerically examined through the problem of inspection maintenance of a structural member group with unevenly scattered fatigue deterioration.