We examine the problem of the determination of the ionization coefficient β from both the theoretical and observational points of view. In the past, theoretical evaluations of β in terms of the relevant scattering cross-sections have used the Massey-Sida formula, which we show to give results which are plainly incorrect. We derive an integral equation for β and compare the results of its application to copper and iron with laboratory simulations. Agreement for the variation of the ionization coefficient with velocity is good. The ionization coefficient has been determined observationally by Verniani & Hawkins from a comparison of radar and visual observations, employing the luminous efficiency τ also obtained observationally by Verniani. However, this determination of τ would appear to be invalidated by fragmentation. There is good evidence that the radiation of cometary meteors is dominated by that of iron in the visual range, and we have accordingly re-analysed the data of Verniani & Hawkins using the luminous efficiency of iron obtained in simulation experiments. However, it is not possible to choose an iron concentration which gives agreement between the determination of the ionization coefficient by this means and its determination from the theoretical equation in terms of either scattering coefficients or simulation methods. The observational ionization coefficients are much lower than predicted by the present theory and we provisionally explain this as a consequence of transfer of charge from the meteoric ion to a molecule of the air. It is now possible for the meteoric atom to be re-ionized, but it is also possible at sufficiently high initial line densities for significant dissociative recombination of the electrons and nitrogen or oxygen to take place. This recombination will not take place in meteor trains simulated in an ionization chamber. We thus conclude that the present theory is limited to faint radio meteors at lower velocities (v≾35 km s−1), for which no significant secondary ionization or recombination will take place. The theoretical results may be approximated by the analytic form β ≃ 9.4 × 10−6 (v − 10)2v0.8, where the velocity v is in km s−1. For visual meteors in the range of about 30 to 60 km s−1, we propose as a reasonable approximation the result we have obtained from the Verniani-Hawkins observational data using simulation results for the luminosity: β = 4.91 × 10−6 υ2.25. At present, however, we are unable to propose estimates of β for slow bright meteors or fast radio meteors.