We report on a sensitive search for mid-infrared molecular hydrogen emission from protoplanetary disks. We observed the Herbig Ae/Be stars UX Ori, HD 34282, HD 100453, HD 101412, HD 104237 and HD 142666, and the T Tauri star HD 319139, and searched for H
2 0-0 S(2) (J=4-2) emission at 12.278 micron and H
2 0-0 S(1) (J=3-1) emission at 17.035 micron with VISIR, ESO-VLT's high-resolution mid-infrared spectrograph. None of the sources present evidence for molecular hydrogen emission at the wavelengths observed. Stringent 3σ upper limits to the integrated line fluxes and the mass of optically thin warm gas (T=150, 300 and 1000K) in the disks are derived. The disks contain less than a few tenths of Jupiter mass of optically thin H
2 gas at 150K, and less than a few Earth masses of optically thin H
2 gas at 300 K and higher temperatures. We compare our results to a Chiang & Goldreich (
1997ApJ...490..368C, CG97) two-layer disk model of masses 0.02M
☉ and 0.11M
☉. The upper limits to the disk's optically thin warm gas mass are smaller than the amount of warm gas in the interior layer of the disk, but they are much larger than the amount of molecular gas expected to be in the surface layer. If the two-layer approximation to the structure of the disk is correct, our non-detections are consistent with the low flux levels expected from the small amount of H
2 gas in the surface layer. We present a calculation of the expected thermal H
2 emission from optically thick disks, assuming a CG97 disk structure, a gas-to-dust ratio of 100 and T
gas = T
dust. We show that the expected H
2 thermal emission fluxes from typical disks around Herbig Ae/Be stars are of the order of 10
–16 to 10
–17erg/s/cm2 for a distance of 140pc. This is much lower than the detection limits of our observations (5x10
–15erg/s/cm2). H
2 emission levels are very sensitive to departures from the thermal coupling between the molecular gas and dust in the surface layer. Additional sources of heating of gas in the disk's surface layer could have a major impact on the expected H
2 disk emission. Our results suggest that in the observed sources the molecular gas and dust in the surface layer have not significantly departed from thermal coupling (T
gas/T
dust<2) and that the gas-to-dust ratio in the surface layer is very likely lower than 1000.