Neuroimaging Meta-Analysis (NMA) is used to synthesize multiple brain imaging studies, a crucial tool for a discipline where N's of 20 or less are typical. Use of NMA has been growing rapidly but suffers from critical limitations of which most users are unaware. Given a selection of neuroimaging studies, e.g., on reward processing, a meta-analysis will produce a picture of coloured regions of significance; a user will use this picture to make spatial and reverse inferences: Spatial, in assertin g that an effect corresponds to a specific anatomical location, and reverse in that a study attribute (i.e. reward processing) is specifically associated with that region. In fact, standard NMA methods cannot provide either: No confidence volumes (3-dimensional confidence intervals) that quantify the uncertainty in the localisation are available, nor can the user conclude that a region is uniquely associated with the selected studies (e.g., reward processing may activate the nucleus accumbens, but so may emotion processing tasks). My fellowship will focus on transforming NMA through development of new methodology and software tools that will address these and other problems. Specific goals include: 1. Development of spatial Bayesian models for NMA data, to deliver true spatial and reverse inferences through a detailed generative statistical model. 2. Detecting, exploring and accommodating inter-study variability with exploratory and confirmatory methods. 3. Correcting publ ication bias (aka the file drawer problem ) & voxel-selection bias (aka the voodoo correlation problem) for NMA inferences. 4. Neuroinformatic tools to facilitate data-sharing and minimize errors and information loss in the process: Data->analysis->publication->meta-analysis.