The aggregate labor share in U.S. manufacturing declined from 62 percentage points (ppt) in 1967 to 41 ppt in 2012. The labor share of the typical U.S. manufacturing establishment, in contrast, rose by over 3 ppt during the same period. Using micro-level data, we document a number of striking facts: (1) there has been a dramatic reallocation of value added to "hyper- productive" (HP) low-labor share establishments, with much more limited reallocation of inputs; (2) HP establishments have only a temporarily lower labor share that rebounds after five to eight years to the level of their peers; (3) selection into HP status has become increasingly correlated with past size; (4) labor share dynamics are driven by revenue total factor labor productivity, not wages or capital intensity; (5) employment has become less responsive to positive technology shocks over time; and (6) HP establishments enjoy a product price premium relative to their peers that causes their high (revenue) productivity. Counterfactual exercises indicate that selection along size rather than shocks or responsiveness to them is the primary driver of the labor share decline.