An experimental apparatus was designed to simulate and study, in laboratory conditions, the flame and the near-field plume stemming from the combustion of an isolated shrub that would be ignited by a surface fire. The main objective is to provide experimental data for the testing of a complete physical model of forest fire behaviour. The laboratory fire source (burner) was made of a cylindrical wire mesh basket filled with a forest fuel ignited at the lower circumference of the basket. A balance recorded the mass loss of the fuel sample and thus enabled to characterise the regime of combustion. Three diameters and two forest fuels were used to obtain different regimes. In addition, a specific device enabled to vary the air supply to the burner. In any case, the regime of combustion was non-steady over the whole duration of the test. Temperatures were measured using type K thermocouples of 50 μ diameter. 28 thermocouples were placed along the vertical axis of the device and along four horizontal lines. In addition, three pairs of thermocouples of which the signal was recorded at a high frequency (200 Hz), were used to determine the upward gas velocity at different heights along the vertical axis thanks to the cross-correlation of thermal fluctuations. Flame heights were obtained from the video recording of each test. The usual scaling laws that enable to relate flame height, temperature profiles and upward gas velocity profiles to steady burner characteristics in the upper part of the flame and in the plume, hold for the maximum flame height, for the maximum temperatures and (less clearly) for the maximum upward velocities observed with the present non-steady burner. But some of the parameters of these laws numerically differ from those obtained in earlier studies of turbulent diffusion flames.