Uranium is the heaviest naturally occurring element and isotope fractionation between 235 U and 238 U is not normally considered significant given the small ∼1% difference in mass. It is therefore usual to assume that 238 U/ 235 U is constant in the terrestrial environment and equal to 137.88. We have developed experimental protocols for the precise measurement of 235 U/ 238 U by multiple-collector ICPMS (MC-ICPMS) and have analyzed a suite of samples formed in a range of low-temperature environments. Using a high-purity 233 U– 236 U double spike to internally monitor the large (percent-level) but essentially constant instrumental mass bias effects that are inherent to plasma source mass spectrometry, we are able to resolve variations in 235 U/ 238 U at the 0.4 epsilon level (2σ; 1 epsilon=1 part in 10,000) on sample sizes comprising 50 ng of uranium. Here we demonstrate sizeable (13 epsilon units) natural variability in 235 U/ 238 U, exceeding the analytical reproducibility by more than an order of magnitude. Compositions that are both isotopically heavier and lighter than our terrestrial standard, by 4 and 9 epsilon units respectively, are observed. The largest excursions are found in speleothem samples. Furthermore, 235 U/ 238 U appears broadly correlated with 234 U/ 238 U in samples showing the most extreme isotopic compositions. The present study investigates the role of abiotic processes in fractionating 235 U from 238 U. Sequential leaching experiments of U-rich minerals indicate that mineral weathering is a possible mechanism by which 235 U can be fractionated from 238 U in groundwaters and incorporated into speleothems. The observed variability in 235 U/ 238 U indicates that uranium isotopes may offer the potential to monitor new reaction pathways, such as those activated during the redox transition between the U(IV) and U(VI) oxidation states. Experiments involving the redox transition of U(VI) to U(IV) in the presence of zero-valent zinc did not produce a resolvable shift in 235 U/ 238 U towards anomalous values, although fractionation need not occur if the reaction is governed by a fast kinetic process. Our observations have a direct impact on the U-series and U–Th–Pb chronometers, when applied to samples formed in low-temperature environments, as these chronometers currently assume an invariant 238 U/ 235 U equal to 137.88.