The authors examine transaction costs associated with algorithmic trading, based on a sample of 2.5 million orders, of which one million are executed via algorithmic means. The data permit a comparison of algorithmic executions with a broader universe of trades, as well as across multiple providers of model-based trading services. Algorithmic trading is found to be a cost-effective technique, based on a measure of implementation shortfall. The superiority of algorithm performance applies only for order sizes up to 10% of average daily volume, however. Algorithmic trading performance relative to a commonly used volume participation benchmark also is quite good, although certainty of outcome declines sharply with the size of the order. A clear link between performance and variability in performance relative to both benchmarks appears to be lacking. Although rough equality across providers is observed on average, this equality of performance breaks down quickly as order size grows.