Abstract: It is important to a pre-warning system for oil-drilling accidents to determine abnormal changes in engineering parameters during the course of oil drilling. A predictive model based on the hierarchical fuzzy system was designed in accordance with high complexities and large numbers of input variables in oil drilling and it could judge the abnormality of parameters by differences between the predicted model output and the actually measured value. A method called ‘fuzzy curve’ was utilized to select and reduce the model inputs from a mass of variables. ANFIS (adaptive neural-fuzzy inference system) was employed to determine the structure of the predictive model and to optimize the parameters of membership functions and fuzzy rules. The introduction of the ‘dynamic fuzzy domain’ concept could solve the problems arising from the slow time-variance of parameter norms in fuzzification. Experimental results indicated that this model showed the superiority both in strong stability of prediction and in high satisfaction for real-time utilization, it responded accurately to varying trends of engineering parameters in oil drilling by taking full advantage of human experience and knowledge. The model was validated with true data of a falling accident for drilling tools and the results suggested that the model could detect abnormal changes in the parameters in time and lay a dependable foundation for the early warning of oil-drilling accidents.