Run charts are widely used in healthcare improvement, but there is little consensus on how to interpret them. The primary aim of this study was to evaluate and compare the diagnostic properties of different sets of run chart rules. A run chart is a line graph of a quality measure over time. The main purpose of the run chart is to detect process improvement or process degradation, which will turn up as non-random patterns in the distribution of data points around the median. Non-random variation may be identified by simple statistical tests including the presence of unusually long runs of data points on one side of the median or if the graph crosses the median unusually few times. However, there is no general agreement on what defines “unusually long” or “unusually few”. Other tests of questionable value are frequently used as well. Three sets of run chart rules (Anhoej, Perla, and Carey rules) have been published in peer reviewed healthcare journals, but these sets differ significantly in their sensitivity and specificity to non-random variation. In this study I investigate the diagnostic values expressed by likelihood ratios of three sets of run chart rules for detection of shifts in process performance using random data series. The study concludes that the Anhoej rules have good diagnostic properties and are superior to the Perla and the Carey rules.