We estimate a forward-looking New Keynesian Phillips Curve (NKPC) for the U.S. using data from the Survey of Professional Forecasters as proxy for expected inflation. We obtain significant and plausible estimates for the structural parameters of the NKPC (the discount factor and the share of firms adjusting prices) independent from whether output or unit labor costs are used as a measure of marginal costs. Survey expectations suggest that the usual identification of expectations exploiting orthogonality of forecast errors with respect to output is severely distorted, which explains why the NKPC estimated with survey data performs much better than under the assumption of rational expectations. We also find that lagged inflation enters the price equation significantly, even when controlling for its ability to predict expectations. This suggests a role for lagged inflation beyond that of capturing non-rationalities in expectations. When estimating a Phillips curve where lagged inflation enters due to price indexation by non-reoptimizing firms, we find that rejection of the coefficient restrictions depends on the measure of marginal costs used.