Face spoofing detection nowadays has attracted attentions regarding the biometrics authentication issue. Inspired by the observation that face spoofing detection is highly relevant with the inherent image quality which also strongly depends on the properties of the capturing devices and conditions, in this paper, we tackle the spoofing detection problem based on a two-stage learning approach. Firstly, we manually cluster the training samples based on the prior knowledge of face sample quality (e.g. camera model), and multiple quality-guided classifiers are trained based on each cluster with extracted image quality assessment (IQA) feature. Subsequently, a regression function is learned by mapping from the IQA scores to the corresponding classifier's parameters, which can be further used for classification. As such, given a new face input for verification, we can predict its classifier's coefficients based on the pre-learned regression model, with which spoofing detection can be effectively achieved. Experimental results show that we achieve significantly better classification performance compared with the strategy that directly applies the IQA features with single classifier.