This paper provides a microstructural explanation for the success of two familiar predictions from technical analysis: 1) trends tend to be reversed at predictable support and resistance levels, and 2) trends gain momentum once predictable support and resistance levels are crossed. ; The explanation is based on a close examination of stop-loss and take-profit orders at a large foreign exchange dealing bank. Take-profit orders tend to reflect price trends, and stop-loss orders tend to intensify trends. The requested execution rates of these orders are strongly clustered at round numbers, which are often used as support and resistance levels. Significantly, there are marked differences between the clustering patterns of stop-loss and take-profit orders, and between the patterns of stop-loss buy and stop-loss sell orders. These differences explain the success of the two predictions.