A new method for measuring two-dimensional granular temperature distributions is presented. This is based on the analysis of the mean squared displacement behaviour of vibro-fluidised grains at short times, using particle position data from digital high-speed video sequences. The performance of this approach is compared with two others: one based on direct computation of the mean squared velocity, and the other fitting a Maxwell-Boltzmann distribution to the measured velocity distributions in the x and the y directions. The new method proved to be the most consistent, particularly at high packing fractions. Typical granular temperature profiles from vibro-fluidised beds are presented; these indicate that the temperature gradient tends to zero at large heights, in contrast to recent Molecular Dynamics (MD) simulations presented by Helal et al. [K. Helal, T. Biben, J.P. Hansen, Physics A 240 (1997) 361.].