Both Vickers and Knoop hardness ( H ), measured at two or more loads in the range of 100–2000 g (most commonly 100 and 500 g) for a variety of dense oxide and non‐oxide materials, covering a range of grain sizes ( G ), including single crystals where possible, were shown to generally be consistent with (often more limited) literature data. Apparently, conflicting trends of H (1) showing either no G dependence, (2) decreasing from single‐crystal or large G values with decreasing G , or (3) having the generally accepted increase with decreasing G are shown to be due to the combination of the limited extent of data and H generally heing determined by two basic trends. These two trends are (a) the normal inverse G (i.e., H–G −1/2 ) dependence at finer G , (b) a variable G minimum at intermediate G , and (c) H increasing with increasing G at larger G (to. single‐crystal values). The H minimum is due to local cracking around the indent (mostly along grain boundaries), generally reaching a maximum effect, e.g., minimum in H , when the indent and grain sizes are similar, and tends to be greater for Vickers vs Knoop indents, higher loads and probably greater grain boundary Impurity, additive contents, and stresses.