A necessary element in the development of predictive procedures for the spread of fires in enclosures is a satisfactory description of the turbulent source. The absence of an established experimental data base inhibits the necessary model development specific to fires. Their buoyant, low source momentum mixing characteristics are quite different from the more widely investigated jet flames and require the application of both particular models and measurement techniques. This paper describes velocity measurements in and around a buoyant flame on a 25 cm diameter porous refractory burner, using LDA with photon correlation signal processing. This technique makes all parts of the flame accessible wilhout recourse to any artificial seeding. The air entrainment region, in which velocities are very low and susceptible to disturbance, is readily, and perhaps uniquely, accessible. Closer to the flame. the large scale inhomogeneities which accompany buoyant flame motion pose important questions of seeding bias as the LDA volume is intermittently engulfed first by hot fuel/product (and therefore higher velocities) and then by cooler ambient air. LDA systems relying on seeding of the fuel stream are particularly susceptible to bias in this regime. Conditional sampling of the velocity, using a thermocouple to trigger the acquisition of data, reveals the scale of differences between these distinctive pockets. On the centre-line, measurements with a variety of techniques are compared and the natural conditioning of the present study, associated with soot concentration and particle size, is used as a diagnostic in an attempt to consolidate all the data.