The detection of moving objects in video sequence is an important task in many video-surveillance systems. As a matter of fact, the output of this very first stage, named background modeling or background subtraction, determine the quality of the rest of pipelines developed for the detection, identification or tracking of persons, objects, etc. Background modeling is sometimes considered either as a trivial operation, carried out by computing a simple difference between the current frame and a single background image (or with the previous frame, etc.), or a mastered technique that do not need any improvement or development nowadays. In the latter case, one cite in general very famous methods such as Gaussian Mixture Models introduced by Stauffer and Grimson in 1999, and think this is sufficient. Unfortunately, these kinds of algorithms are limited in outdoor environments, when used in long-term surveillance applications, because of many uncontrolled and damaging events: global variation of luminance, shadows of objects, bad weather, camera tilts, etc. Since this is a key-point of video-surveillance applications, background subtraction has become a popular topic, and many techniques have been proposed since the 90's. In BMC (Background Models Challenge), we propose a new benchmark composed of almost 30 synthetic and real video sequences. Thanks to these data-sets, we are able to propose very complex situations, in various surveillance contexts (human activities or traffic for example). We have also developed a free software (BMC Wizard) that computes relevant criteria that evaluate statistical, signal and structural information from a background subtraction algorithm.