Four pigeons responded on a two-component multiple token-reinforcement schedule, in which tokens were produced according to a random-interval 30-sec schedule and exchanged according to a variable-ratio 4 schedule in both components. To assess the effects of contingent token loss, tokens were removed after every second response (i.e., fixed-ratio 2 loss) in one of the components. Response rates were selectively lower in the loss components relative to baseline (no-loss) conditions, as well as to the within-condition no-loss components. Response rates were decreased to a greater degree in the presence of tokens than in their absence. To control for the effects of changes in the density of token and food reinforcement, two parts consisted of additional conditions where food density and token loss were yoked to those in a previous loss condition. In the yoked-food condition, tokens were produced as usual in both components, but the overall density of food reinforcement in one of the components was yoked to that obtained during a previous token-loss condition. In the yoked token-loss condition, tokens were removed during one component of the multiple schedule at a rate that approximately matched the obtained rate of loss from a previous token-loss condition. Response rates in these yoked components were less affected than those in comparable loss components, despite similar densities of token, exchange, and food reinforcement. On the whole, the results support the conclusion that contingent token loss serves as an effective punisher with pigeons. [ABSTRACT FROM AUTHOR]