Froth visual features are closely related to flotation performance, and accurate extraction of froth visual features through machine vision allows improved control and optimization of the flotation process. However, the conventional froth features extracted from single two-dimensional (2D) images are inadequate to fully characterize froth features due to the loss of depth information. In this paper, we present a new method of stereo visual feature extraction on single 2D images for significantly improving froth characterization. First, a froth-specific reconstruction approach is proposed to derive 3D representation of froth through a model of defocus blur and illumination. Then, the froth stereo features are extracted based on the 3D representation. The new 3D reconstruction approach is validated by the effect of recovered 3D scenes under different conditions. Compared to the conventional methods, the froth stereo features have less fluctuation and improved stability. Finally, recognition of stereofeature based working conditions is investigated for the industrial flotation process. Our results show that the froth stereo features can adequately characterize the data with respect to their separability, regardless of the classifier used.