A review of previous research on surface nucleation in glasses demonstrates that these are mostly qualitative and that strong discrepancies exist regarding the nucleation mechanism. In this article, the surface crystallization kinetics of several glasses - a Na20.3Ca0.6SiO2 (devitrite), a non-stoichiometric devitrite, and two commercial soda-lime-silica (a float and a microscope slide) glasses - were determined in a wide range of temperatures and time. An analysis of the average number of crystals per unit area, N(s), crystal growth rates and viscosity data indicates that the surface nucleation rates are very high and that crystallization arises from a fixed number of special sites, N(s). The number of crystals nucleated strongly depends on the surface condition (e.g., fire polished versus mechanically polished or as-received; clean versus dirty), on the chemical composition of the parent glass, and also on the nature of the crystallizing phase. However, N(s) does not depend on time or temperature. The experimental evidence indicates that the surfaces 'per se' do not alter the thermodynamic barrier for nucleation (the interfacial energy or the chemical potential). The enhanced nucleation rates at the external surfaces are the result of the catalytic effect of some (unknown) solid impurity particles and faster surface diffusion rates.