Dust determines the temperature structure of protoplanetary disks. However, dust temperature determinations almost invariably rely on a complex modeling of the Spectral Energy Distribution. We attempt a direct determination of the temperature of large grains emitting at mm wavelengths.} We observe the edge-on dust disk of the Flying Saucer, which appears in silhouette against the CO J=2-1 emission from a background molecular cloud in $\rho$ Oph. The combination of velocity gradients due to the Keplerian rotation of the disk and intensity variations in the CO background as a function of velocity allows us to directly measure the %absorbing dust temperature. The dust opacity can then be derived from the emitted continuum radiation. The dust disk absorbs the radiation from the CO clouds at several velocities. We derive very low dust temperatures, 5 to 7 K at radii around 100 au, which is much lower than most model predictions. The dust optical depth is $> 0.2$ at 230 GHz, and the scale height at 100 au is at least 8 au (best fit 13 au). However, the dust disk is very flat (flaring index -0.35), which is indicative of dust settling in the outer parts.