Facial expression recognition (FER) is a crucial technology and a challenging task for human computer interaction. Previous method has been using different feature and classification method for FER and use basic method for testing. In this paper, we used the best feature descriptor for FER by empirically evaluating two feature descriptors, namely Facial Landmarks, and Binary Robust Independent Elementary Features (BRIEF) descriptors. We examine each feature descriptor by considering one classification method such as Support Vector Machine (SVM) with three unique facial expression recognition datasets. In addition to test accuracies, we present confusion matrixes of FER. We also analyze the effect of using this feature and image resolutions on FER performance. Our study indicates that Facial Landmarks descriptor works the best for FER to run on smart phone. The experimental results demonstrate that the proposed facial expression recognition on mobile phone is successful and it gives up to 96.27 % recognition accuracy.