The Castleton karst (Derbyshire, UK) is at first sight typical of many karst systems worldwide. There are inputs of allogenic water from streams that sink at 15 points and inputs of autogenic water fed by rain and snow falling onto the soil-covered limestone outcrop which is pitted by dolines. Water output takes place from three springs that feed a surface stream, the Peakshole Water. Between sinks and springs there are 45 caves with a total surveyed length of over 38km. The longest is the Peak-Speedwell system which is upstream of the springs and has around 50 inlet streams of which 20 flow from permanently water-filled ‘sumps’. Cave divers have explored over 1000m of these sumps one of which descends to a depth of at least 76m below dive-base. Over 50 water-tracing experiments have revealed the broad outline of the underground hydrology but there is a great deal of internal complexity that is yet to be fully understood. From July 2012 to April 2015 Water depth and temperature were measured at 1-minute or 2-minute resolution in the two main inlet sumps in Speedwell Cavern, at the three springs, and in the Peakshole Water. The data reveal a greater level of complexity than is present in any published data set that I am aware of. The sumps, several hundred metres apart and at different elevations, exhibit both flow switching (the bulk of the flow from the allogenic sinks sometimes entering via one and at other times via the other) and nonlinearity (rapid changes of depth and temperature but with no consistent periodicity). Depth changes can be as rapid as +20cm and -18cm per minute. From the inlet sumps vadose streams flow to a downstream sump that connects with the springs. Some of the complexity is retained in the spring output response (depth changes of up to +8cm and -8cm per minute) and in the stream-discharge downstream of the springs. For example, in one event the discharge fell from 1429L/s to 822L/s over a 54 minute period and then rose back up to 1429L/s over the following 142 minutes. The data pose three questions for karst hydrogeologists and modellers: (1) how can an apparently simple system (when seen in plan) develop such a complex response to recharge; (2) are there any tools that would allow deductions to be made on the system geometry purely based on the output and (3) are there any tools that could provide a reasonable prediction of output.