Word production has been investigated using magnetoencephalography [MEG] for 20 years. We perform a review and meta-analysis of MEG studies of word production published in the last 20 years (1994-2014). We assess significant findings with the goals of gauging spatial and temporal resolution, evaluating the between-studies consistency of the cortical response, and the reliability of cognitive inferences made from these findings. We use the author's own attribution of activity to function and current serial and interactive models to scaffold our analysis. Despite current data trending towards seriality in the dynamics of component functions of speech, no strong claims are warranted regarding the serial or parallel nature of cognitive processes. We suggest that future studies might employ standardised protocols and analyses, and report their results more comprehensively to make inferring cognitive function and relating MEG data to psycholinguistic models more efficient.