The aim of the paper is to determine (endogenously) whether the volatility of the US output growth rate has changed since the late 1940s. By applying the discrete wavelet transform to the annualized quarter-to-quarter output growth series, we test the homogeneity of the variance on a scale-by-scale basis. A version of the Normalized and Centered Cumulative Sum of Squares test, adapted to wavelets, leads us to reject the null of constant variance in the two levels of decomposition of the highest resolution and to locate a single break in 1982. The economic implications are explored.