Using both conventional imaging and confocal systems it is possible to collect large data sets rapidly. For a number of reasons the exploration and analysis of these data sets can be far more time consuming than the original experiments. First, the large quantity of data takes a significant amount of time to load into memory. Second, the stepwise iteration of regions of interest (ROI) through the data set can be computationally expensive. Third, with many commercial software packages the analysis involves making intermediate files of identical size (for example when using a spatial filter) that further expand the data set and cannot be easily modified once made. Here I present a compiled Visual Basic (Microsoft) program that overcomes some of these problems. The program allows for ROI analysis of large data sets. It implements a simple set of ideas to produce significant speed and flexibility gains when compared with most commercial software. The software uses some principles of the kanban business model (Sugimori et al., 1977) more commonly known as ‘just-in-time’ (JIT). The main benefit of this model is that large amounts of unnecessary ‘stock’ are not held. Instead only that which is instantaneously required is obtained. In computing terms this means that when a file is opened only the first image is displayed – this can be done almost instantly. Then, when a ROI is drawn only the data from that region is read from the file. Furthermore, if the data are to be plotted, only the frames that can be plotted on the screen are analysed. For example, an image series lasting 10 min acquired at 512×512 pixel resolution at 25 frames s^-1 using an 8-bit model will generate at least 3.9 GB of data (15,000 frames). Typically, such images might cover an area of 100 µm x 100 µm (pixel size ~0.2 µm). A circular region of interest of, for instance a cell body might be 10 µm in diameter and would therefore comprise ~2000 pixels. This represents ~0.8% of the total number of pixels in the frame. An initial overview of the mean intensity for that region, on a standard 1024×768 monitor, requires data to be calculated from every 1/15^th frame (1000 data points). Thus, the data actually required is only 2 MB in total. This is ~0.05% (1/2000^th) of the complete data set. If the analysis process is rate limited by the data transfer rate – which is frequently the case – this kanban process could yield a 2000-times faster analysis. Of course, the analysis of smaller regions in longer time-series would produce greater gains whereas for larger regions in shorter time-series smaller gains would be expected. Optimization of the kanban process, by the underlying memory cache of the operating system, means that once data are extracted for a given region subsequent modifications to the analysis occur without large amounts of data transfer. Thus, the kanban-driven acquisition occurs with the same short latency as if the whole data set had been loaded into memory. Furthermore, operations on the data such as spatial and temporal filtering are applied as necessary to the data ‘on the fly’ – this is again a kanban principle. This means that the filtering can easily be adjusted and data replotted without the need for the production of intervening files. Part of the JIT process is that certain operations are predicted, and those not associated with a time overhead are performed in advance. For example, numerical ROI data are available on the clipboard immediately after plotting so that they can be exported to other analysis software. The software also implements the simple idea of non-contiguous pixel-based ROIs. This allows pixels from anywhere in the image to be selected and combined into a single functional region. Currently, data files from Leica (LIF), Zeiss (LSM), TIF, BMP and JPG files can be analysed. The program allows background subtraction, single and dual wavelength ratios, sub-pixel resolution image shifting (based upon a linear-interpolation algorithim), temporal and spatial smoothing, basic curve fitting and calibration of ionic data. The program will function with large (>2 GB data sets) on a wide range of Windows machines even with very modest (256 MB) memory specifications taking data across Ethernet networks.