Han's (2009, 2013) selective fossilization hypothesis (SFH) claims that L1 markedness and L2 input robustness determine the fossilizability (and learnability) of an L2 feature. To test the validity of the model, a pseudo-longitudinal study was designed in which the errors in the argumentative essays of 52 Iranian EFL learners were identified and categorized based on a researcher-developed error taxonomy. Next, the learners were provided with implicit and explicit corrective feedback on those errors to see if there existed any errors that would persist despite learners' motivation and the pedagogical intervention to eliminate them from their writing. ANOVA results revealed that the errors in the "pronoun," "word order," "passive voice" and "possessive categories" persisted in the written output of the participants. A sub-classification of errors in the pronoun and possessive categories showed that deletion and redundant addition of subject pronouns, lack of agreement between pronouns and their antecedents, vague or ambiguous pronoun references in the pronoun category, and a wrong use of the apostrophe (') or apostrophe + s ('s) with regular plurals in the possessive category proved most resistant to correction, pointing to their tendency towards becoming fossilized. That tendency, however, could not be accounted for by the SFH.