The classical biological control is a viable, effective strategy and approach to control alien invasive weeds as manifested by numerous successes in the long history of classical biological control. However, introduction of nonnative biological control agents from their native region can pose potential direct or indirect ecological risks to nontarget native organisms, which calls for necessary alertness. There exist flaws in the conventional practices in assessment of potential risks of biological control agents, which are manifested by: (1) host specificity testing depends mostly on physiological (fundamental) host range testing in the lab, but less on ecological (realized) host range testing which is a prediction of host use under the range of physical and biotic conditions in the new environment; (2) the physiological host range testing mostly relies on the ontogeny in the lab, but less on behavioral and genetic characteristics; (3) the risk assessment pays more attention to nontarget plants of economic importance but less on organisms in natural ecosystems. To reduce the risks in conventional practices of biological weed control, it is suggested that: (1) retrospective ecological studies of released natural enemies should be encouraged, so as to provide theoretical support for the risk assessment of releasing biological control agents; (2) for the application of biological control approach to management of alien invasive weeds, it is recommended that adoption of “harmful until proven safe” approach based on the Precautionary Principle could be exercised to avoid unforeseen negative effects; （3） more emphasis on assessment of ecological effects should be laid on the risk assessment of biological control agents.