A field experiment was conducted to examine the factors in threat appeal responsible for maintenance of crime-prevention behavior. At four public bicycle parking lots, 256 riders with variable receiver vulnerability characteristics were encouraged to use an extra bicycle lock. They were randomly presented with one of three threat messages featuring victims of bicycle theft (identifiable victim, statistical victim, or control), followed by either high- or low-efficacy preventative-messages. After extra locks were installed on their bicycles, participants’ use of the lock was observed five times within 28 days after the intervention. A mixed-effect generalized linear model revealed that vulnerability of the participants increased the use of the lock immediately after the intervention. Meanwhile, highly vulnerable participants who were presented with an identifiable victim and high-efficacy messages decreased their use of the lock significantly compared to low-vulnerability participants and those who were presented with the low-efficacy message. The result implies that threat appeal strategies differ depending on receiver vulnerability and the type of preventative behavior.