The aim of this study is to address the managementof urinary problems by detecting changes in the volume of urinewithin the human bladder using low cost, low power, wearable UltraWideband (UWB) sensors and signal processing techniques. Thepaper describes experiments on the classi¯cation of six three-layerdielectrically representative bladder phantoms, mimicking a rangeof muscle and bladder wall-to-wall distances. The process involvesthe illumination of the bladder with a UWB pulse. Due to thedielectric contrast between urine and bladder wall tissue at microwavefrequencies, an electromagnetic re°ection is generated at both theanterior and posterior bladder wall. These re°ections are recorded,the salient features are extracted and processed by a classi¯cationalgorithm to estimate the volume of urine present in the bladder.To evaluate the prototype system, a number of physical bladderphantoms were constructed, each mimicking bladders of di®erentvolumes. Principal Component Analysis (PCA) was applied and theprocessed features were classi¯ed by a K-Nearest Neighbour learningalgorithm to estimate the state of the bladder (small, medium or full).The paper describes the bladder phantom prototype systems and theexperimental setup. Results illustrate detection of phantom bladderstates with an accuracy of up to 91%.