This research investigates two factors influencing the ability of tree-ring data to provide accurate 14C calibration information: the fitness and rigor of the statistical model used to combine the data into a curve; and the accuracy, precision and reproducibility of the component 14C data sets. It presents a new Bayesian spline method for calibration curve construction and tests it on extant and new Southern Hemisphere (SH) data sets (also examining their dendrochronology and pretreatment) for the post-Little Ice Age (LIA) interval AD 1500–1950. The new method of construction allows calculation of component data offsets, permitting identification of laboratory and geographic biases. Application of the new method to the 10 suitable SH 14C data sets suggests that individual offset ranges for component data sets appear to be in the region of ± 10 yr. Data sets with individual offsets larger than this need to be carefully assessed before selection for calibration purposes. We identify a potential geographical offset associated with the Southern Ocean (high latitude) Campbell Island data. We test the new methodology for wiggle-matching short tree-ring sequences and use an OxCal simulation to assess the likely precision obtainable by wiggle-matching in the post-LIA interval.