This paper presents a new technique to extract the light invariant local feature for facial expression recognition. It is not only robust to monotonic gray-scale changes caused by light variations but also very simple to perform which makes it possible for analyzing images in challenging real-time settings. The local feature for a pixel is computed by finding the direction of the neighboring of the pixel with the particular rank in term of its gray scale value among all the neighboring pixels. When eight neighboring pixels are considered, the direction of the neighboring pixel with the second minima of the gray scale intensity can yield the best performance for the facial expression recognition in our experiment. The facial expression classification in the experiment was performed using a support vector machine on CK+ dataset The average recognition rate achieved is 90.1 ± 3.8%, which is better than other previous local feature based methods for facial expression analysis. The experimental results do show that the proposed feature extraction technique is fast, accurate and efficient for facial expression recognition.