This pseudo-longitudinal study adopted typical-error and corrective feedback approaches to investigating interlanguage fossilization. The errors in the argumentative essays of 76 Iranian EFL learners within and across three proficiency levels were identified and classified using the model proposed by Gass and Selinker (1994). The learners were first provided with implicit and then explicit feedback to see if the two feedback types would improve written production, and if there were errors that persisted. The results indicated that word choice, plural, word form and article “the” were the most frequent types of errors, with word choice topping the list. The results also showed that, although providing learners with feedback could lead to a reduction in errors, explicit feedback was a more effective strategy. However, the t-test results demonstrated that the short-term impact of feedback could not be sustained over time, a finding congruent with those of Truscott (2007) who questioned the efficacy of error correction. Moreover, the findings revealed that the error categories of pronoun, word order, passive and possessive were likely candidates of fossilization because, although for all the other error types implicit feedback could be beneficial, these showed resistance to correction and needed explicit feedback to be eliminated.