Wages are rigid if they vary «less than they should» because economic mechanisms, mental attitudes or institutional constraints prevent them from changing, especially decreasing. Empirical tests of downward rigidities are not robust. They are based on ad-hoc assumptions regarding the shape that the distribution of wage variations would have if the wage rigidities were absent. Under standard assumptions, a large proportion of zero wage variations due to a small proportion of decreases is interpreted as indicating the existence of rigidities. An analysis made for France based on four sources of information on full-time wages in the market sector suggests that the application of these tests to household survey data greatly overestimates wage rigidity. We use reputedly reliable sources to show that wages were highly variable in France in the late 1990s. Every year, wages fell for 20% to 30% of employees. Zero variations from one year to the next were rare. However, basic wage variations, excluding bonuses, displayed characteristics compatible with the presence of rigidities. Wage variability is therefore posited to be partly due to the variability of bonuses. The analysis suggests that another form of wage rigidities exists in the shape of a partial adjustment of wages to productivity shocks affecting firms. This adjustment is more marked in the case of a positive shock than in the case of a negative shock.