Variability in L2 learners’ and heritage speakers’ ability to produce/comprehend gender morphology has been variously attributed to lack of access to UG (Franceschina 2001), performance limitations (White et al. 2004), or underspecification of features (McCarthy 2007). To diagnose the source of this variability, task effects and systematicity of errors have been appealed to. In the present study, variability in the ability of heritage speakers of Modern Greek to deal with gender morphology is examined using a production task and a written recognition task. The results reveal no differences in task performance and a systematic use of the default gender. This suggests that variability in gender morphology lies in the underspecification of morphological features that predicts unidirectional patterns of errors across the board.