A 0.9 km-reach of Uvas Creek, California, was reconstructed as a sinuous, meandering channel in November 1995. In February 1996, this new channel washed out. We reviewed project documents to determine the basis for the project design and conducted our own historical geomorphological study to understand the processes operating in the catchment and project reach. The project was designed using a popular stream classification system, based on which the designers assumed that a "C4" channel (a meandering gravel-bed channel) would be stable at the site. Our historical geomorphological analysis showed that the reach had been braided historically, typical of streams draining the Franciscan Formation in the California Coast Ranges, with episodic flows and high sand and gravel transport. After the project washed out, Uvas Creek reestablished an irregular, braided sand-and-gravel channel, although the channel here was narrower than it had been historically, probably due to such factors as incision caused by gravel mining. Our study casts doubt on several assumptions common in many stream restoration projects: that channel stability is always an appropriate goal; that channel forms are determined by flows with return periods of about 1.5 years; that a channel classification system is an easy, appropriate basis for channel design; and that a new channel form can be imposed without addressing the processes that determine channel form.