Weed control decisions are modeled in an endogenous risk framework where a producer invests in self-insurance and self-protection to reduce the severity of a realized pest infestation, or reduce the likelihood the infestation occurs. Self-insurance and self-protection are risk-reducing technologies that capture both the type and quantity of herbicides used. We supply conditions to unambiguously sign the effects of an increase in the probability of application or effectiveness failure and increased application or effectiveness uncertainty on optimal herbicide choices. If self-protection and self-insurance are stochastic substitutes, non-point source pollution policies targeted to reduce herbicide loadings can increase the use of more persistent herbicides. Policies that decrease loadings by reducing total mass may induce a substitution to herbicides more damaging or more likely to be transported to sensitive areas.