A Barro-type economic growth model is estimated for the 50 states in the U.S. using data for three decades beginning in 1960. Frontier estimation techniques are used to test for the presence of state-specific shocks to economic growth that are independent of the usual, normally-distributed random errors. We find that large, positive shocks to growth occur during the period 1960-90. Our results indicate that the error term structure assumed each other OLS may not be appropriate for modeling economic growth.