We make a new model for pattern failure, which is the pattern collapse and bridging of resist patterns of 1:1 lines and spaces (L/S) exposed as a focus-exposure matrix, to explain and predict the process window of the pattern failure. It is found that the conventional model cannot be fitted to the experimental pass/fail data. Instead of and , we select the critical dimension (CD) and normalized image log slope (NILS) as the model input. The new CD-NILS model corresponds well to the experimental pass/fail data. Good correspondence is assumed to be due to the properly selected model input. Pattern collapse, which occurs during the drying of the water at the rinse of the resist patterns, is expected to be accelerated by the smaller line CD and the larger line width roughness (LWR) due to smaller NILS. Pattern bridging, which occurs during resist development, is expected to be accelerated by the larger line CD and the larger LWR. The CD-NILS model predicts the process window precisely when a new process condition (a new illumination in this case) is adopted. It suggests that the CD-NILS model is a powerful methodology for predicting the process window to optimize the process condition and optimize the lithography design.