We present an alternative to growth accounting à la Solow, on the same set of variables, that provides a metric for labor-saving technical change ('λ') and capital-saving technical change ('µ'). These two components are identified through the variations of the factor shares, which we assume to reflect marginal productivities. We run our algorithm using BEA data from 1948 to 2015, and compare the predictive power of our time series of (λ_t,µ_t) with the one of the Solow residual. Through simple regressions, we find: (i) that λ and µ are as good predictors of the growth rate of GDP per capita as the Solow residual, and (ii) that λ and µ, together with capital accumulation, are strong predictors of the variation of the factor shares, while the Solow residual is not. We conclude that a bi-dimensional representation of productivity has a stronger empirical relevance than the usual linear representation; however the former carries some different theoretical properties than the latter – notably on the consequences of capital accumulation.