Three main problems organize the analysis of analogical change in morphology: (1) the relation between leveling and extension; (2) directionality; and (3) selectivity. The selectivity problem arises in cases where an analogical change such as paradigm leveling affects some but not other seemingly equally eligible items: What factors determine this analogical selectivity? More specifically, what sorts of items resist otherwise expected analogical change? At least three specific types of analogy-resistant item have been identified in the liteature: relatively frequent items; items whose alternations are in stressed syllables; and items with multiple morphophonological alternations. The last pattern has been called "Paul's Principle" (Paul 1880); all three patterns concern properties of the analogy-resistant item itself. Based on evidence from English and Latin, I will identify a new and somewhat different pattern of resistance to analogy, which I call "paradigmatic heterogeneity": items in morphophonologically more heterogeneous paradigms may resist analogical change, even if the locus of heterogeneity lies outside the area targeted by the change. I will suggest that Paul's Principle may be a special case of this more general pattern, that it may provide evidence for paradigms as objects in morphological analysis, and that it may cast light on the overall cause of analogical change in morphology.