The current article represents a methodological proposal. It seeks to address the question of how one might recognize a discovery as a discovery without knowing in advance what is available to be discovered. We propose a solution and demonstrate it using data from a study previously reported by J. Roschelle (1992). Roschelle investigated 2 students' developing understandings of certain abstract features of Newtonian mechanics while working within a computer-based microworld, the Envisioning Machine. We employ an approach we term "discovery as occasioned production" to reexamine his data. Such an approach proceeds stepwise from the identification of some matter discovered, working backward to see just where that matter entered the conversation and then, finally, tracing from that point forward to illuminate how the proposal for a possible discovery was ultimately transformed into a discovery achieved. The notion of "evident vagueness," borrowed from H. Garfinkel, M. Lynch, and D. Livingston's (1981) account of the discovery of an optical pulsar, emerges as an important feature of our analysis. Following H. Garfinkel (2002), we present our findings as a "tutorial problem" and offer a suggestion for how a program of practice studies in the learning sciences might be pursued. (Contains 7 figures, 20 footnotes, and 9 excerpts.)