Proposes a new model for determining when to pause in Japanese. Our model, which simulates the left-to-right processing of human speech production, incorporates the two main factors of pausing: phrase length and limited syntactic information. By taking this approach, our model can explain several types of pausing phenomena in Japanese, which cannot be explained by the naive pausing model. To prove the validity of our model, we implemented it in a text-to-speech conversion system. Through a listening test, we compared the proposed model with a naive pausing model mainly based on phrase length. Results show that the proposed model performs well.