Since manual error detection of morpho-syntactic errors of English writing requires lots of time and resources, automation of error detection is essential in both Computer-Assisted Language Learning and English learning studies. This approach aims at automatic detection of morpho-syntactic errors of English writing using association rule mining, which needs three steps of procedures. As the first step, we generate association rules based on the refined data. Second, we statistically verify the generated rules. Third, we testify the verified rules on the test data. Previous studies have focused on either word errors based on the language models using large corpora, or the systems very specific to the complex grammatical theories. Meanwhile, this study uses relatively small amount of data. We used the Apriori algorithm for the rule mining task. Since rules generated by the algorithm can contains lots of noise to be reduced, we apply statistical machine learning methods using correlation coefficient and cosine similarity. This process sifts valid mal-rules for the automatic detection tasks from lots of noise.