To improve quality of personalized recommendation and simplify the preference setup in generating recommendation rules, the characteristics of the association rule for personalized recommendation are discussed, the concepts of recommendation nonblank metric, a new recommendation metric, 1-support frequent itemset and k-maximal association rule are defined, and the idea of getting k-maximal association rule from 1-support frequent itemset is proposed. Moreover, an association rule mining algorithm based on the idea is designed, which is suitable for different sliding window depths. The theoretic analysis and experiment results on the algorithm show that the method has maximal nonblank, higher precision and F-measure of recommendation, and simplifies the preference setup of thresholds in mining rules effectively.