Steel is one of the important materials in the industry. Steel may have defects in the production process that can affect the steel products. Therefore, the detection of steel surface defects is an important process to control the quality of steel products. An efficient steel surface detection process is carried out by automating steel images taken using a camera. We use an anchor-free model FoveaBox. FoveaBox is an accurate and flexible model for detecting objects and has a simple architecture. This study uses the NEU-DET dataset consists of six types of steel surface defects, namely crazing, inclusion, patches, pitted surface, rolled-in scale, and scratches, each with a total of 300 data.. The test results on the system show that the method used has a good detection performance with a mean average precision of 0.834 or 83.4% at a learning rate of 0.001, Optimizer SGD, sigma 0.6, and the number of epochs 24. This detection method can detect steel surface defects. This detection method can effectively detect steel surface defects with similar foreground and background characteristics. With an accuracy threshold of 80%, the method used in this study has an adequate precision value.