This study explored learning and generalization of parts of the Russian case-marking paradigm, an inflecting-fusional system in which affixes simultaneously mark several grammatical features (case, gender, number, animacy). In Experiment 1, adult English speakers (N = 43) were exposed to nouns with transparent gender marking in the nominative case and learned to inflect nouns for two oblique cases. Experiment 2 (N = 44) used nouns that in the nominative case were morphophonologically nontransparent (i.e., arbitrary) with respect to the underlying gender categories. Learning and generalization of case-marking inflections were better in Experiment 1 than Experiment 2, but far worse than would be expected if learners had extracted the underlying rules. Measures of proficiency in other languages with transparently marked grammatical gender categories, fluid intelligence, and verbal working memory capacity were predictive of learners' performance in both experiments, although their relative impact differed somewhat depending on the transparency of the system. These findings suggest that adult learners benefit from regularity when learning morphological patterns but do not necessarily generate rules, as the complexity of morphological variation in inflecting-fusional systems inevitably requires a considerable degree of item-based learning.