The physiological significance of calcium sparks in cardiac, skeletal and smooth muscle cells has received a great deal of attention in recent years. Such studies rely upon the accurate detection and analysis of brief, localised release events in order to quantify changes in spark frequency and properties under different conditions. In order to adequately characterise the time course of sparks, images are often obtained using confocal laser scanning microscopy in linescan mode. The manual analysis of this data is time consuming and prone to user bias, because the small size of many events and relatively low signal-to-noise ratio make it difficult to discern small amplitude sparks from noise artifacts. The main source of noise in these images is photon noise, which follows a Poisson distribution and entails that the noise variance is larger when the background fluorescence is increased. Previously published automated spark detection algorithms are limited in at least one of two related ways: (1) sparks are assumed to occur from a constant baseline, and (2) the noise is assumed to be Gaussian, with a fixed variance throughout the image. These algorithms are therefore inappropriate for reliably identifying sparks in images containing multiple cells with differing baseline fluorescence levels, or occurring on top of global calcium elevations. Both situations arise in images of calcium signaling events in the smooth muscle of retinal arteriolar segments, as described by Curtis et al. (2004). We have developed new software that overcomes these issues by using a wavelet-based variance stabilisation technique to automatically adapt spark detection to changes in baseline fluorescence. In addition to facilitating the analysis of images for which no automated algorithm currently exists, preliminary tests indicate that the use of a more accurate noise model means that our algorithm can also offer improved detection accuracy in any linescan containing sparks, particularly at low signal-to-noise ratios.