For automatic pilling evaluation of textiles, the depth information is one of the most critical and effective features in extracting pills from fabric image. Laser-scanning techniques are often used for acquiring 3D depth images. However, due to the high-cost and low-efficiency of Laser-scanning system, researchers have found it unsuitable for fabric analysis. This paper illustrates a new approach for acquiring the depth image used to extract pills by introducing the method of Depth From Focus (DFF). This approach firstly captures a sequence of images of the same view at different focal positions under the automatic optical microscope. Then the best-focused position (z) of each pixel(x, y) was determined by choosing the layer of image declaring the max sharpness and formed the depth image. This paper proposed a new sharpness-evaluation criterion which was based on the variance of gradients. Afterwards, a few basic points indicating the background area was selected from the depth image, and then the depth coordinates (x, y, z) at these basic points were used to calculate a predicted background plane. Via the background plane, pills above the background were extracted. A fabric sample with a single fiber upon it was presented to illustrate the process and result of the approach.