We develop and assess new time series measures of economic sentiment based on computational text analysis of economic and financial newspaper articles from January 1980 to April 2015. The text analysis is based on predictive models estimated using machine learning techniques from Kanjoya. We analyze four alternative news sentiment indexes. We find that the news sentiment indexes correlate strongly with contemporaneous business cycle indicators. We also find that innovations to news sentiment predict future economic activity. Furthermore, in most cases, the news sentiment measures outperform the University of Michigan and Conference board measures in predicting the federal funds rate, consumption, employment, inflation, industrial production, and the S&P500. For some of these economic outcomes, there is evidence that the news sentiment measures have significant predictive power even after conditioning on these survey-based measures.