Abstract A fire occurrence module was developed for CFES-IAM Version 2, a stochastic simulation model of initial attack on wildfires in California. The module is designed to generate annual sequences of fire start times that are consistent with local fire history. A three-stage approach was employed: (1) For each day of a simulated fire season, a random draw from a Bernoulli distribution is used to determine if any fires occur; (2) if any do occur, a random draw from a geometric multiplicity distribution determines theft number; (3) ignition times for each fire are then randomly drawn from a time of day (beta or Poisson) distribution. This approach and specific distributional forms were selected after analysis of historical fire records from California's Sierra foothills and Central Valley. Fire sequences generated with the module appear to capture historical patterns with respect to diurnal distribution, interfire times, and total number of fires per year. For. Sci. 34(4):948-959.