We studied the importance of selection and constraint in determining the limits of adaptive radiation and the consequences of adaptive radiation in an experimental system. We propagated four replicate lines of the bacterium Pseudomonas fluorescens derived from a single ancestral clone in 95 environments, where growth was limited by the availability of a single carbon source for 1,000 generations. We then assayed the growth of the ancestral clone and the evolved lines in all 95 environments. Evolved lines increased their performance in almost every selection environment and invaded 70% of the novel environments as a direct response to selection. Direct responses tended to be larger in environments where growth was initially poor. Although evolved lines lost the ability to grow on about three substrates that their ancestor could readily grow on, the correlated response to selection was, on average, positive. The correlated response allowed all of our evolved populations to expand their niches and to occupy collectively the remaining novel habitats. This is inconsistent with classical theories of niche evolution. In the most extreme cases, adaptation occurred through "roundabout selection": lineages became adapted to an environment through selection in another environment but not through selection in the environment itself. Our results indicate that mutation accumulation by neutral drift was responsible for generating the majority of costs of adaptation.