This study compares the relationship between official crime rates in census tracts and resident perceptions of crime. Using a unique data set that links household-level data from the American Housing Survey metro samples over 25 years (1976-1999) with official crime rate data for census tracts in selected cities during selected years, this study finds that tract violent crime is the strongest predictor of residents' perception of crime. This standardized coefficient was 0.71 on average over the seven waves. Models simultaneously taking into account both violent and property crime found a consistently strong positive effect for violent crime but a consistently negative effect for property crime. Among types of violent crime, robbery and aggravated assault have the strongest effect on the perception of crime in the tract. Burglary showed a stronger effect on perceptions of crime in the 1970s but a steadily weakening effect since then. There was little evidence that the racial/ethnic composition of the tract affected these perceptions. (Contains 4 tables and 13 notes.)