Automatic opinion/sentiment analysis is essential for analysing large amounts of text as well as audio/video data communicated by users. This analysis provides highly valuable information to companies, government and other entities, who want to understand the likes, dislikes and feedback of the users and people in general. Opinion/Sentiment analysis can follow a classification approach or perform a detailed aspect level analysis. In this paper, we address a problem in between these two, that of segmentation and classification of opinions in text. We propose a recurrent neural network model with bi-directional LSTM-RNN, to perform joint segmentation and classification of opinions. We introduce a novel method to train neural networks for segmentation tasks. With experiments on a dataset built from the standard RT movie review dataset, we demonstrate the effectiveness of our proposed model. Proposed model gives promising results on opinion segmentation, and can be extended to general sequence segmentation tasks.