In this project, I extended a convolutional neural network (CNN) for multi-class animal face classification using the 20-class LHI-Animal-Faces dataset. I modified the training pipeline to support multi-class output, incorporated validation loss tracking, and used it to determine the optimal model. I evaluated performance by adding batch normalization and dropout layers, and performed hyperparameter tuning with different optimizers, learning rates, and batch sizes. The best test accuracy of 77% was achieved using the Adam optimizer, a 1e-3 learning rate with scheduling, dropout after the final convolution layer, and a batch size of 16.