Context: The CoRoT satellite searches for planets by applying the transit method, monitoring up to 12 000 stars in the galactic plane for 150 days in each observing run. This search is contaminated by a large fraction of false positives, caused by different eclipsing binary configurations that might be confused with a transiting planet. Aims: We evaluate the rates and nature of false positives in the CoRoT exoplanets search and compare our results with semiempirical predictions. Methods: We consider the detected binary and planet candidates in the first three extended CoRoT runs, and classify the results of the follow-up observations completed to verify their planetary nature. We group the follow-up results into undiluted binaries, diluted binaries, and planets and compare their abundances with predictions from the literature. Results: 83% of the initial detections are classified as false positives using only the CoRoT light-curves, the remaining 17% require follow-up observations. Finally, 12% of the candidates in the follow-up program are planets. The shape of the overall distribution of the false positive rate follows previous predictions, except for candidates with transit depths below about 0.4%. For candidates with transit depths in the range from 0.1-0.4%, CoRoT detections are nearly complete, and this difference from predictions is probably real and dominated by a lower than expected abundance of diluted eclipsing binaries. The CoRoT space mission, launched on December 27th 2006, has been developed and is operated by CNES, with the contribution of Austria, Belgium, Brazil , ESA (RSSD and Science Programme), Germany and Spain.