The temporal and spatial variability of crescentic sandbars is analyzed with hourly long-term (months) video observations collected at four barred sites and are qualitatively compared to the temporal and spatial variability predicted by hypotheses underpinning existing approaches and models for crescentic bar formation (edge-wave template model, linear stability analysis, and nonlinear models). The observations, coming from the single barred beaches at Duck (North Carolina, USA) and Miyazaki (Kyushu, Japan), and from the double-barred beaches at the northern Gold Coast (Queensland, Australia) and Noordwijk (Netherlands), show that crescentic sandbar wavelength and amplitude variations over space and time are very common. For instance, at any moment in time, the wavelength of the smallest and longest crescentic bar can differ by a factor of 2. Temporal changes in wavelength and amplitude result from merging and splitting of individual crescents, causing the final configuration of a crescentic sandbar system to be very different from the initial configuration. The Gold Coast data indicate that these intrinsically nonlinear interactions are an attempt of the crescentic bar system to self-organize into a more uniform pattern, as splitting is usually confined to the longest crescentic bar observed, whereas merging usually combines the smallest crescentic bars into a longer bar. The observed spatial and temporal crescentic bar behavior contrasts qualitatively with behavior predicted from the edge-wave template model and implies that the predictive skill of linear stability models is limited. Nonlinear models are potentially better suited for a comparison against these field observations; several suggestions to improve these models, and hence to facilitate a data-model comparison, are made.