PURPOSE: to compare the severity of phonological disorder from the perceptual analysis of three distinct groups of judges. METHODS: the research consisted of two samples, one to be assessed with 30 individuals with phonological disorder with ages between 4:1 to 7:11, and another sample that comprises the judges (five speech therapists, five laypeople and five mothers). We evaluated the spontaneous speech of the children through a three logical sequence narrative to be assessed by the judges, accompanied by a grid where to register the severity level. For data analysis, the mode of 90 pieces of narratives was carried through, making possible the statistical analysis of the data based on Kappa Concordance Analysis. RESULTS: the analysis showed higher concordance level among the groups of judges in the extremes of the possibilities of assessment (mild and severe), being more difficult the judgment of mild-moderate and moderate-severe. CONCLUSION: we observed that it was easier for judges to act in the identification and judgment of those subjects with severe (higher) and mild (lower) speech severities. The laypeople have more difficulty to judge accurately the severity of phonological disorder than mothers and speech therapists.