Objectives The study evaluated the interpretation of, and preferences for, numerical information on side-effect incidence when presented in three different formats. Methods It used a controlled design, with participants allocated at random to receive one of the three formats. Participants were recruited via a pop-up window on the CancerHelp UK website. The sample comprised 129 website users, of whom 96% were women and 86% had cancer, who received frequency information on four side effects of tamoxifen, using one of three risk expressions (percentages, e.g. ‘affects 25% of people’; frequencies, e.g. ‘affects 1 in 4 people’; combined, e.g. ‘affects 1 in 4 people (25%)’). They then interpreted information on tamoxifen and its effect on health, and estimates of side-effect frequency, and then stated a preference from the three risk expression formats. Key ﬁndings The results showed that the three formats did not inﬂuence participants’ ratings of the information or their side-effect estimates. However, more than half (53%) the participants preferred the combined (frequency and percentage) format. In conclusion, a combined risk expression format performed no worse than percentages or frequencies alone and was preferred by a majority. Conclusions The three risk expression formats did not differ in their effect on participants’ interpretations. However, the preferred format was the combined (frequency and percentage) risk expression.