The problem of invasive species has reignited interest in biological control as a management tool.Classical biological control involves deliberate release of exotic natural enemies into new environments in an attempt to limit the density of an invasive species. Persistent, sustained limitation of invasive species by coevolved natural enemies is a seductive concept. Evidence now suggests, however, that biological control through the release of natural enemies can carry unanticipated ecological risks. There have been ecological side effects of distributing a deliberately introduced weevil ( Rhinocyllus conicus) and an adventitious weevil ( Larinus planus) for the biological control of exotic thistles. Both weevils have had major direct effects on key population-growth parameters of native thistles, and R. conicus has had an indirect effect on the interaction between a thistle and a native insect. These findings led us to review how ecological risk is evaluated, and to ask whether pre-release tests can predict the types of ecological effects documented. We conclude that, when done thoroughly, the tests used can determine host specificity by identifying physiological host range, but the usual tests cannot be relied upon to predict the ecological host range or impact on populations of less-preferred but accepted native species. Our data provide support for suggestions that the behavioral and developmental data now taken need to be supplemented with additional data on population parameters to better predict field-host use, population growth, interaction strengths, and ecological outcomes for native species that are potential hosts.