Motorcycle deaths accounted for more than half of total traffic fatalities in Taiwan in 2002. This study uses the police-reported crash data from Taiwan between 1999 and 2001 to estimate the effectiveness of helmets, simultaneously taking into account of sample selection bias. Sample selection arises because helmet usage will affect the probability of death or injury, which in turn influences whether a crash is included in the data. The results show that sample selection does not seriously bias the estimate of helmet effectiveness and helmets reduce the probability of death in a crash by 40%, which is higher than what was previously found. Without helmets, the number of motorcyclists killed in 2001 would have jumped by 51%. The estimated proportion of helmeted motorcyclists has increased from 71 to 78% between 1999 and 2001, suggesting that helmet use is rising after the implementation of mandatory helmet law in 1997. Also, helmets significantly reduce the likelihood of head and neck injuries in a crash by 53%, and lead to a 71% reduction in the probability of death caused by head and neck injuries.