This paper addresses several important issues related to crime. First, we construct a violent crime index taking into account seven different types of crimes. We use an aggregator function to define a crime index that attaches crime-specific weights which can be interpreted as severity of each crime. These weights are estimated econometrically along with other parameters in the model thereby avoiding the problems associated with equally or arbitrary weighted aggregate crime index. Second, we utilize the aggregate crime index function to determine the impact of socio-economic variables on the overall (aggregated) crime, and further decompose them into crime-specific components. Third, in specifying the crime index we allow the possibility that crimes may be underreported and estimate crime underreporting using the stochastic frontier modeling approach. We use district level data from India for the census years 1981, 1991 and 2001. Our results fail to support the equally weighted crime index model and provide evidence of substantial underreporting.