Using an online or inline capillary rheometer as a tool of rheology measurement would come into the ends pressure drop problem. In order to derive the actual pressure drop of the capillary, another capillary with the same diameter and different length is needed (according to Bagley correction) but would result in a more complex mechanism. In this study, a neural network approach is proposed to estimate the ends pressure drops in an online capillary rheometer. The back propagation learning algorithm is used for network training. The shear rate, the die pressure, and the ratio of diameters of the reservoir to the capillary are taken as the neural network inputs, and the ends pressure drop is taken as the output. Two hundred of training sets that are made from a laboratory capillary rheometer are used for network training. The trained neural network can be consequently applied to real-time assessment of the ends pressure drops in the online capillary rheometer. It is concluded that using the proposed method for calculating the ends pressure drop is effective. Besides, the simplicity of the mechanism provides good portability for both online polymer characterization and quality control in processing.