During language acquisition, children must learn when to generalize a pattern – applying it broadly and to new words (‘add –ed’ in English) – and when to restrict generalization, storing the pattern only with specific lexical items. One effort to quantify the conditions for generalization, the Tolerance Principle, accurately predicts children’s generalizations in corpus-based studies. This principle hypothesizes that a general rule will be formed when it is computationally more efficient than storing lexical forms individually. Here we test the principle in an artificial language of 9 nonsense nouns. As predicted, children exposed to 5 regular forms and 4 exceptions generalized, applying the regular form to 100% of novel test words. Children exposed to 3 regular forms and 6 exceptions did not extend the rule, even though the token frequency of the regular form was still high. The Tolerance Principle thus captures a basic principle of generalization in rule formation.