We use data from the manuscript censuses of manufacturing for 1850, 1860, 1870, and 1880 to study the dispersion of average monthly wages across establishments. We find a marked increased in wage inequality over the period, an increase that cannot be explained by biases in the data or changes in census enumeration procedures. Based on log wage regressions on establishment characteristics (for example, size, capital intensity, use of steam power, and so on) we compute a decomposition of the change in wage inequality between 1850 and 1880. The decomposition reveals that changes in "wage structure" - the regression coefficients and the standard error of the residuals - largely offset each: changes in the coefficients produced a reduction in wage inequality, while residual inequality increased. Most of the rise in wage inequality can be attributed to an increased concentration of employment in large establishments, which paid relatively low wages. We present indirect evidence that the negative effect of size on wages reflected differences in skill composition: workforces in large establishments were less skilled than in small establishments.