In a practical multiphase batch process, the durations of batches are probably different, and at the same time, the length of each phase may also vary from batch to batch. This paper proposes an effective method to handle the uneven length problem in multiphase batch processes, which is also useful when the number of training batches is limited. To address the data nonlinearity, two nonlinear modeling techniques are introduced. For online monitoring and quality prediction, a key step is how to locate the current data sample to the specific phase, on which basis an appropriate model can be employed. In this paper, a phase discrimination model for data localization in different phases that is based on the support vector data description model is developed. For performance evaluation of the proposed method, detailed illustrations of a typical multiphase batch process are provided.