Probabilistic models for wind loads are developed and used to estimate the properties of the responses of linear systems. This analysis involves four steps. First, along-wind, across-wind forces and torque are represented as polynomials of turbulence fluctuations and wake excitations which are assumed to be stationary Gaussian processes. Second, we provide two types of models, namely, the empirical model and the mathematical model, for the second-moment properties of the turbulence fluctuations and wake excitations so that the probability law of the wind loads is characterized completely. Proposed models are then calibrated to the experimental observations. Third, the mathematical model provides an efficient method to estimate the response properties relative to Monte Carlo simulation. The responses are modeled by translation processes that match the target second-moment properties and marginal distributions of the responses. Fourth, the response properties which are of interest, e.g., the mean rates at which the responses exit the safe set, are calculated from the resulting translation processes. We illustrate this procedure by one numerical example.