A series of replicate experiments, involving analysis of homogenized wood and identical tree-ring sections, suggests that, for this laboratory at least, the ¹⁴C counting error in radiocarbon dating quantifies only part of the total variability of measurement. In summary, it is found that the normally quoted error on a ¹⁴C date should be multiplied by a factor of two to give a more reasonable minimum estimate of error. Assessment of the various alternative methods of ¹⁴C age calibration reveals that inclusion of a more realistic error term for the date to be calibrated is more vital than either the method of deriving an age calibration system or the precision of the calibration chronology. Nevertheless, statistical methods of non-parametric regression estimation, involving cross-validation to assess smoothness, provide both objectivity and error consideration in calibration curve derivation and in chronology-matching and are therefore recommended. -- AATA