Facial feature extraction is an important aspect in facial image perception system. And it is also a prerequisite in animation system for generating a given person's 3D-face image. In this paper, a coarse-to-fine facial feature extraction strategy is presented based on facial texture distribution and deformable template, using the pre-result of a multi-level face detection, which aims at solving such problems as the searching highly depending on the initial parameters and time-consuming that deformable template algorithm often suffers from. In proposed strategy, firstly, the center of the two irises is localized making use of the valley and frequency characteristics in the two eye regions. Then integral projection is used to localize the coarse position of the mouth and the nose. Secondly, some key feature points about these organs are estimated. Finally, according to these feature points, good initial parameters for the pre-defined templates are given and an optimal algorithm based on greedy algorithm and multi-epoch cycle is used to search for the minimum solution. Experiments indicate that the implementation of the proposed strategy is with good performance in both speed and accuracy.