This study explores how learners generalize grammatical categories such as noun gender. Adult native English speakers with no prior knowledge of Russian (N = 47, ages 17–55 years) were trained to categorize Russian masculine and feminine diminutive nouns according to gender. The training set was morphophonologically homogeneous due to similarities induced by the diminutive suffixes. Generalization was tested using simplex and diminutive nouns with trained or novel stems. Age, familiarity with other second languages, nonverbal intelligence, and verbal working memory capacity were used to predict performance. Compared to trained diminutives, learners showed equally high accuracy in gender categorization of transparent simplex nouns with trained stems and diminutive nouns with novel stems. Accuracy dropped significantly for nontransparent simplex nouns with trained stems and transparent simplex nouns with novel stems. Verbal working memory capacity predicted gender categorization of nontransparent nouns and vocabulary recall, whereas age, nonverbal intelligence, and prior experience with languages with transparent gender categories affected performance in nouns containing at least some degree of gender transparency. The findings indicate that pattern-based and memory-based learning processes coexist in adult learners. However, age affected performance on pattern-based processes but not memory-based processes like rote memorization of vocabulary and gender categorization of nontransparent nouns. This finding is interpreted in the context of proposals concerning the role of declarative and procedural memory in L2 learning.