This paper discusses new bridge models for short-term forecasting of French quarterly GDP growth. The only data used are from business surveys in French manufacturing, services, and construction. We consider two alternative methods. The first relies on the general-to-specific (GETS) algorithm applied to blocks of randomly selected variables (Hendry and Krolzig, 2005) ; the other relies on the combination method popularized by Stock and Watson (2004). We conduct in-sample and out-of-sample assessments of both methods using recursive and rolling regressions. We show that the forecast based on an automatic regression-model selection (GETS) performs better, and that extending the database to business surveys in the service and construction sect...