The properties of exchange rate forecasts are investigated, with a data set encompassing a broad cross section of currencies. Over the entire sample, expectations appear to be biased. This result is robust to the possibility of random measurement error in the survey measures. There appear to be statistically significant differences in the degree of bias in subgroupings of the data: (i) the bias is lower for the high-inflation countries; (ii) the bias is greater for the major currencies studied in earlier papers; and (iii) the bias is also greater for the EMS currencies.