Despite the popularity of business training among policy makers, the use of business training has faced increasing skepticism. This is, in part, fueled by the fact that most of the first wave of randomized experiments in developing countries could not detect statistically significant impacts of training on firms' profits or sales. This paper revisits and reassesses the evidence for whether small business training works, incorporating the results of more recent studies. A meta-analysis of these estimates suggests that training increases profits and sales on average by 5 to 10 percent. The author argues that this is in line with what is optimistic to expect given the relatively short length of most training programs, and the expected return on investment from the cost of such training. However, impacts of this magnitude are too small for most experiments to detect statistically. Emerging evidence is provided on five approaches for improving the effectiveness of traditional training by incorporating gender, kaizen methods, localization and mentoring, heuristics, and psychology. Training programs that incorporate these elements appear to deliver improvements over traditional training programs on average, although with considerable variation. Given that training delivers some benefits for firms, the challenge is then how to deliver a quality program on a cost-effective basis at a much larger scale. Three possible approaches to scaling up training are discussed: using the market, using technology, or targeting and funneling firms.