This paper describes the results both of a rational analysis of discovery learning and of an analysis of think aloud data. These data were gathered while subjects leam the effects of different nominal factors in a simulated domain. The reported analyses are conducted to guide the modeling of discovery learning in Act-R.