The dissociation of hydrogen at a tungsten filament has been studied in the temperature range 1200 to 1800°K and at pressures between 10 -2 and 10 -6 mm. The results obtained differ from those found in the literature and it is shown that this is because the surfaces employed by previous workers were contaminated, very probably by vapours derived from tap grease. At temperatures below 1400°K and at pressures exceeding 10 -6 mm, the rate of atomization of hydrogen is given by the expression, v a (atoms cm -2 s -1 ) = 18 x 10 24 ( P H 2 mm)½ exp (52600/ RT ). The theory of absolute reaction rates is applied to the two possible mechanisms, namely W – H = W + H and H 2 + W = W – H + H ; under the conditions considered, equilibrium between adsorbed and gaseous hydrogen is maintained throughout the reaction. Reasons are given for rejecting the second mechanism, which was the one favoured in the past. From a consideration of both the atomization and recombination reactions, it is demonstrated that the first equation is applicable and, further, that the adsorbed atoms have full translatory freedom on the surface at the temperature of reaction. The observation that molecular and atomic hydrogen leave the filament in their equilibrium ratio, as determined by the temperature of the filament and the pressure of hydrogen, is shown to be compatible only with this description of the reaction. At temperatures in the region of 1800°K, the rate of reaction ceases to be proportional to √ P H 2 at relatively high pressures and has become linearly dependent on P H 2 at pressures less than 10 -6 mm. This behaviour is discussed quantitatively in terms similar to those employed for the low-temperature reaction, but now there no longer exists an equilibrium between adsorbed and gaseous hydrogen. A study of the atomization on surfaces exposed to oxygen indicated that the adsorbed oxygen layer is rapidly removed by exposure to hydrogen at 1200°K and that, thereafter, the dissociation proceeds at the same velocity as on a clean surface. The reaction on a carbided surface occurred at a rather slower rate than obtained with a clean surface, but the activation energy remained unaffected. The nature of the contamination responsible for the low activation energies of about 45 kcal/mole reported by previous workers has thus not been identified.