In response to an incoming fire alarm, someone must decide how many and which engine and ladder companies (firefighters and their apparatus) to dispatch to the scene. Traditional dispatching policies assume that all of the designated companies are available at the time the alarm is received. These policies do not consider the workload imposed on firefighters, and do not work well at the high alarm rates now characteristic of parts of large cities. Our procedure, designed for use in New York City's computed-aided Management Information and Control System, makes good initial dispatch decisions at all alarm rates. It uses response times to serious fires to measure performance and recognizes that the dispatcher has incomplete information about the seriousness of the incident when the decision is made. A simulation comparison of our procedure with the traditional policy was made using actual incidents from July 1972. Our procedure reduced the average second engine and second ladder response times to serious fires by 25 to 45 seconds, while keeping total workload essentially unchanged. This reduction, which is about 10--15% of the five and a half minutes obtained for the traditional policy, results primarily from our procedure's deciding how many to send based on historical information on the chance that an alarm from a given location at a given time of day is serious. For example, the fraction of fires in occupied structures getting an initial second engine rose from 65% with the traditional policy to 85% with our procedure, although both policies sent an initial second engine to the same fraction of all incidents. Our procedure also reduced the number of relocations of engine companies and ladder companies substantially. The approach we used should be valuable in the design of computer-aided dispatching systems in other cities. In particular, others may find it helpful to review the way in which the objective function is developed, the way particular aspects of the dispatch problem are treated, the provision of several parameters for tuning behavior, and features of the simulation testing.government services: fire, heuristic models, simulation: use in policy analysis