I examine the relationship between firm size and skill composition, using microdata from Australian workers and firms between 2011 and 2020. While larger firms generally employ higher shares of high-skill workers, all firms tend to hire relatively more low-skill workers as they grow. To explain these patterns, I develop a model that distinguishes total factor productivity (TFP) from skill-biased productivity (SBP), in which firms choose their scale and workforce composition. I validate the model mechanisms using evidence from a payroll tax policy change in South Australia, demonstrating how firms adjust both scale and skill mix in response to cost changes. I use a quantitative model of the Australian labor market to explore the implications of shifts in aggregate skill composition on the distributions of firm size and earnings inequality. I find that an 11 percentage point increase in the educated share of the workforce decreases equilibrium skill premia within all firms. Despite this, the aggregate skill premium increases, because firms that employ higher shares of high-skill workers raise earnings levels more for all their workers. The increased share of educated workers also leads to employment gains of 3\% in the largest firms and an aggregate reallocation of 1\% of workers to the largest firms. The results highlight how accounting for employment composition decisions by firms is crucial for understanding observed patterns of worker skill distribution and earnings across firms.