Biological control projects for spotted, diffuse and squarrose knapweeds and yellow starthistle started 30 to 40 years ago. How can we shorten the time needed to discover, evaluate, approve and distribute new agents? How can we assess a priori which agents are likely to have the most efficacy? Host specificity testing has a good record of predicting risk of agents directly damaging nontarget plants, but how can we minimize undesirable indirect impacts on nontarget plants and animals? Experiences with the knapweed and yellow starthistle projects provide some examples of what we can do better and what we should avoid.