The use of value-added modeling (VAM) in school accountability is expanding. However, trying to decide how to embrace VAM can be rather nettlesome. Some experts claim it is "too unreliable," causes "more harm than good," and has "a big margin for error," while other experts assert VAM is "imperfect, but useful" and provides "valuable feedback." This article attempts to parse these statements by exploring the underlying statistical assumptions of VAM, the reliability of VAM's estimates, and the validity of the inferences commonly made based on the estimates of VAM. It then goes on to discuss the perverse incentives, unintended consequences, and gaming that might accompany the misuse of VAM. The article concludes that while, in many cases, VAM may be preferable to other commonly used measurement modes, it should never be used as the sole indicator of teacher effectiveness. Rather, it should just be a piece of a larger accountability system. (Contains 23 notes.)