Study data were gathered in 1977 and came from 24 police agencies in 3 metropolitan areas: Rochester, N.Y.; St. Louis, Mo.; and Tampa-St. Petersburg, Fla. Data came from interviews with 1,435 officers; telephone interviews with 12,019 randomly selected citizens in 60 neighborhoods; and observations of 5,682 encounters between citizens and police. The analysis considered 7 indicators of police attitudes, 26 neighborhood variables, and a series of dummy variables describing characteristics of suspects and complainants. The specific situation was the most important factor determining the police officer's variable. Two neighborhood characteristics also increased the likelihood of arrest: a high rate of serious crime and a high transiency rate. Two attitudinal variables were also important: police perceptions that citizens are likely to use police services reduced the chance of arrest, whereas perceptions of high crime risks to citizens increased the likelihood of arrest. Officers need to question their tendency to be more punitive toward nonwhite suspects than toward white suspects. Arrest may not be the most effective response to a situation. Tables, appended lists, and 71 references.