Based on self-reported reference wages of Japanese union workers we provide new insights on the relative income hypothesis. The availability of a direct measure of peers' wages overcomes one of main limitations of previous empirical tests of the relative income hypothesis, i.e the sub jectivity of the denition of reference group. Our data allows us to study the relationship between happiness and reference wages using the following measures of reference group wages: i) based on predicted individual wages; ii) based on cell averages from internal, as well as, iii) external data; iv) based on colleagues' average wage; and compare the results based on these proxies with the results based on the "true", self-reported, reference wage. We show that the proxies that have been used in the literature do not introduce a simple classical measurement error and that the bias can go in both directions. We propose a simple IV strategy when the elicited reference wage is not available that does not solve the bias but at least delivers an upper-bound of the true eect. We also deal with the possible endogeneity of self-reported reference wage due to an underlying pessimism that is correlated with life and job satisfaction.