I show that the US tax multiplier depends on the direction of the tax change. The tax multiplier is significantly larger (in absolute value) for tax hikes than for tax cuts – regardless of whether I identify tax shocks via (i) the narrative approach or (ii) sign restrictions. The tax hike multiplier is strongly pro-cyclical, i.e., substantially larger in expansions. Variation in the tax cut multiplier over the business cycle is milder and statistically insignificant. A simple business cycle model with downward nominal wage rigidities can explain these results.