Research on the detection reliability of ultrasonic infrared imaging is of great importance. Probability of detection for different defects under given test conditions is workhorse statistical technique for the detection reliability assessment. In this paper, a series of 45 steel containing fatigue cracks were tested. Experimental results show that the increase of the crack size will enhance the crack heating response, and the relationship between the logarithmic form of measured response and the crack size appears to be linear. Based on the above statistic characteristics, the linear regression analysis and Delta method were adopted to estimate the probability of detection and its confidence interval. The research conclusion will provide a quantitative basis for detection reliability in ultrasonic infrared imaging.