The long-term efficacy of written grammar feedback (WGF) has escaped rigorous empirical investigation, partly due to its strong intuitive appeal. The few studies that have investigated the long-term effect of grammar correction have failed to find any positive long-term effects for WGF (Truscott, 2007). Nevertheless, a number of recent studies have reported a positive effect for WGF. The present study investigated the longterm effect of selective grammar feedback on a reasonably complex feature of the English grammar, attempting to shed some light on the factors that may explain the conflicting results of previous studies. A group of low intermediate EFL learners (N=22) participated in this study. They received WGF on the use of articles and a limited number of various other grammatical categories. The results showed that selective WGF can produce large short-term gains for functionally complex grammatical features, but that it may prove to be detrimental in the long run. The study also found that learners tend to avoid the grammatical feature on which they have received corrective feedback.The theoretical and practical implications are discussed.