The paper explores how light fields captured byplenoptic cameras can increase the performance of face landmarkdetection. The idea is to exploit light fields geometrical constraintsto correct the position of points detected by classical facelandmark detectors. These geometric constraints are used toenforce landmark points angular coherency across the differentviews of the light field, and by doing so to correct the positionsof the landmarks on all views. The corrected landmark pointsare compared with ground-truth manual annotations of a set of400 images corresponding to the central views of 400 light fieldsof faces with different pose and expression.