Features extracted from convolutional neural networks are becoming increasingly popular for machine learning tasks. The usual procedure is to get a conv-net pre-trained on any dataset for the same application as the current objective and to use it to get features of the current dataset to do classification. In this paper, study the performance of features extracted from a pre-trained CNN that was trained on ImageNet and fine tuned on PASCAL for the problem of object classification. We attempt to use various methods of classification to get the best possible performance. We measure performance in terms of time taken as well as accuracy to determine which techniques are best for the purpose of object classification.