Streaks are one of the most common defects in electrophotographic printers, and dramatically affect print quality. Researchers have developed methods to detect streaks. Then, using the detection result helps us to diagnose issues of the printer and discover broken components of the electrophotographic printer. In previous work, the streak detection methods are based on a particular printer or particular streak defects, such as Intermediate Transfer Belt (ITB) or Organic Photoconductor (OPC) drum streak. In this paper, we design a Block Window Method to pre-test the images with streak defects. It is based on the local ΔE value in a block window and works for different kinds of streaks. After using the Block Window Method, the detection result includes small streaks or noise defects that are too localized for humans to see. We use the logistic regression algorithm to classify the real visible streaks and small invisible streaks. This process will improve the accuracy of the detection result. After the classification, we can get the streak detection result, which is significant for extracting the feature vector of the streak defects in the test image. Then, we can use the feature vector to classify different streak defects.