When children reduce onset clusters to singletons, a common pattern is for the least sonorous member of the adult cluster to be produced. Within OPTIMALITY THEORY (Prince & Smolensky, 1993), this pattern has been accounted for in terms of a fixed ranking of onset constraints that evaluate a segment's degree of sonority, whereby onset glides violate the highest ranked constraint, and onset stops the lowest. Not all children follow the sonority pattern, however. In this paper, we apply two fundamental principles of optimality theory to yield predictions about other children's cluster reduction patterns. The first principle is that of FACTORIAL TYPOLOGY, according to which all rankings of constraints should yield possible languages. To produce the sonority pattern, all conflicting constraints must rank beneath the onset sonority constraints. If they rank above the onset sonority constraints, these other constraints will force deviations from the sonority pattern. In this paper, we show how divergences from the sonority pattern are caused by three well-motivated conflicting constraints: *FRICATIVE, *DORSAL, and MAX-LABIAL. This is documented in the speech of two normally developing children (about 1;6–2;3) and a child with a phonological delay (3;8). The second principle we appeal to is that of EMERGENT CONSTRAINT ACTIVITY, according to which the effects of violated constraints can be observed when higher ranked conflicting constraints are not at issue. We show that even when the onset sonority constraints are outranked by the conflicting constraints, under the right circumstances the sonority pattern does emerge in the forms produced by these children.