This document presents a multi-layer convolutional neural network (MLCNN) for classifying Twitter sentiment on an ordinal scale of five points (Highly Positive, Positive, Neutral, Negative, Highly Negative). The MLCNN uses different filter sizes and pooling techniques to capture the complexity of ordinal classification. It outperforms previous state-of-the-art models on the SemEval 2016 Twitter sentiment dataset, achieving a MAEM score of 0.617 using various filter sizes and average pooling. The MLCNN is able to automatically learn features and representations from word embeddings to perform Twitter sentiment analysis, without extensive feature engineering.