KNN algorithm is a commonly used algorithm in the assessment of optical image key event. However, classical KNN algorithm always makes conclusion unreasonable, because it only concerns about the number of candidate cases, neglecting of candidate cases’ private characters. To solve this problem, an improved KNN algorithm was proposed, which focused on the private characters of candidate cases. This paper argued that, the distance between candidate cases and target case, and the probability distribution of the candidate cases, both had important influence on last conclusion of target case. The test results showed that, the KNN algorithm proposed was more accurate than classical KNN algorithm, and the membership in proposed KNN algorithm represented degree of success or failure, which were more practical and more reasonable in the engineering practice.