Current concepts of flame propagation in premixed, turbulent gas streams are examined. This leads to the conclusion that the link between theory and experiment is entirely inadequate and incapable of improvement by existing methods. A series of new methods is implemented in an attempt to short-circuit the chain of hypothesis and experiment which has hampered the identification of dubious steps. Methods of introducing uniform turbulence at relatively slow flows and improvements in light sources allow analysis of the approach flow by photographing particles illuminated by an interrupted Tyndall beam. Three new optical deflexion methods are used to give a measure of the randomness of flame-front orientation, of the time-mean structure of the flame and of the instantaneous shape of the corrugated front. It is found that this corrugated surface propagates at a velocity considerably in excess of the normal laminar burning velocity. Quantitative analysis of the frequency of `peaks' and `valleys' on the surface, together with comparative data from the apex of laminar flames, suggests an explanation in terms of the effects of curvature and, secondarily, of the influence of small-scale turbulence.