Experiments in the gas phase usually involve averaging observables over a random molecular axis alignment distribution. This deleterious averaging limits insights gained by probes of molecular dynamics, but can be overcome by prealigning molecular axes using laser-alignment methods. However, the transformation from the laboratory frame to the molecular frame of reference requires quantitative knowledge of the axis alignment distribution. The latter is often hard to obtain directly from experimental data, particularly for polyatomic molecules. Here we describe a general maximum-likelihood classification procedure for non-adiabatic numerical alignment simulations with free parameters that employs experimental data from an alignment-dependent probe. This method delivers (i) the most probable molecular frame angular dependence of the probe, and (ii) the most likely laboratory frame axis alignment distribution of the sample, each with a confidence interval. This procedure was recently used for studies of angle- and channel-resolved strong field ionization of 1,3-butadiene in the molecular frame [Mikosch et al., Phys. Rev. Lett. 110, 023004 (2013)], used here as an illustrative example.