Studies were conducted in 1994 and 1995 to examine the effects of a range of action thresholds for managing Bemisia tabaci (Gennadius) Biotype B (=B. argentifolii Bellows & Perring) with insecticides in cotton on populations of arthropod predators in Imperial Valley, CA, and Maricopa, AZ. Application of insecticides significantly reduced population densities of spiders, Geocoris punctipes (Say), G. pallens (Stål), Orius tristicolor (White), Nabis alternatus Parshley, Zelus renardii Kolenati, Hippodamia convergens Guérin-Méneville, Spanogonicus albofasciatus (Reuter), Drapetis sp., and Chrysoperla carnea Stephens in one or both years and sites compared with untreated controls. Use of higher B. tabaci thresholds conserved some species and groups relative to lower thresholds. Stepwise regression analyses indicated that reductions in predator populations were generally influenced more strongly by the timing of the first insecticide application than by the total number of sprays necessary to maintain suppression of the pest below any given action threshold. A predation index, which weights the importance of each predator species based on their known frequency of predation on B. tabaci and another key pest, Pectinophora gossypiella (Saunders), was developed and analyzed. Patterns were similar to results based on changes in abundance alone, but the index generally revealed less severe effects of insecticides on overall predator function. The current action threshold for conventional insecticidal control of B. tabaci in Arizona and southern California is five adults per leaf. Results here suggest that predator conservation may be enhanced by raising the initial threshold to delay the first application or initially using more selective materials such as insect growth regulators.