The cognitive engineering principle suggests that the unsafe behaviors of construction workers are associated with numerous attributes, and 23 task demands and 12 capability attributes have been proposed in the construction worker’s behavior model (CWBM). Two models utilizing Logistic Regression (LR) and Artificial Neural Networks (ANN) were developed as accident prediction models, and the forecasting efficiencies of these two models were investigated. Robustness of these models was proven by verification. The results provide a basis for designing an in-depth study on the cognitive attributes influencing workers’ behaviors and expanding the choice of analysis techniques.