The present study attempted to assess the reasons behind the inter-observer variation in grading oral epithelial dysplasia (OED). Three oral pathologists and one general pathologist examined 68 histological slides of OED lesions of variable grade for scoring the presence of each individual characteristic of the architecture and cytology changes that were established by the 2005 WHO classification. The assigned features in each case were correlated with clinical outcomes to understand which features are more commonly associated with malignant transformation. Interestingly, for all individual characteristics, the pairwise inter-examiner and group kappa values ranged from poor to moderate. It appeared that for each characteristic separately there was much dissension. Despite these observations, comparing these data with that from our previous paper on the same slides showed that the inter-observer agreement on the degree of dysplasia either by using the new binary system of "low-risk" or "high-risk" or by using the 2005 WHO classification turned out to be better than the agreement on the individual characteristics of architecture and cytology changes. Certain features show significant association with the clinical outcomes. In the discussion, some explanations to help understanding the sources of variation in grading OED are put forward. In conclusion, grading dysplasia is not an exact science and pathologists are doing their best to reach optimal results. Improvement in the standard of the histopathology reporting of OED lesions could be achieved by consideration of several issues. Of these, there is need for a universal definition of the architectural and cytological features that are the basis of any OED grading process. A minimum dataset for reporting OED lesions should be set up. Also, the use of a consensus scoring process between two or more observers should be encouraged as this would improve inter-observer agreement.