Machine learning has been used to recognize emotions in faces, typically by looking for 8 different emotional states (neutral, happy, sad, surprise, fear, disgust, anger and contempt). We consider two approaches: feature recognition based on facial landmarks and deep learning on all pixels; each produced 58% overall accuracy. However, they produced different results on different images and thus we propose a new meta-classifier combining these approaches. It produces far better results with 77% accuracy