Referential Cohesion Analysis with the supplementary non-cohesive taxonomy was used here for the analysis of two corpora. Reliability of the analysis was ensured by the analysis of one RA by two coders (reliability was measured the same way as in Stage 2), with a higher percentage of inter-coder agreement (89%) than at the earlier stage of the analysis (85.2%), which means that this analysis supports the assumption that considering non-cohesive items enhances the reliability of the analysis. The overlap between the author’s two analyses of the same text was 96%. Differences in the intra-coder analyses were mainly due to ambiguities (with most of the differences appearing in the categorization of exophoric items), and to assigning different presupposed elements to a few items in the cohesive table. The other coder also reported difficulty in identifying the subcategories of non-cohesive items. In fact, the specific sub-categories are useful in identifying where specific items belong, for pedagogical purposes – or otherwise, more efficient Referential Cohesion Analysis – the use of the five main categories is recommended – keeping the subcategories and examples for clarity. Problematic items were marked in the analyses and will again be discussed in connection with the errors found in the MA papers (Section 7.6.5).