Some aspects of the self‐trapped propagation of beams of strong electromagnetic radiation are discussed. A comparison is made between a variational approach and the commonly used paraxial ray approximation, demonstrating how the latter gives an erroneous result for the nonlinear wavenumber.