Abstract Rapid growth of popular and lucrative catch-and-release marine sportfisheries worldwide has highlighted the need for reliable weight estimation methods for use in fishing tournaments, pursuit of fishing records, and to support scientific research that allows captured fish to be released alive. This paper describes new methods to predict weight from measurements of body size of game fish. Here we also evaluated efficacy of a widely used historical model and compared this to several new statistical and analytical weight estimation models that we developed herein. We applied these candidate models to a unique morphometric data set on Atlantic tarpon (Megalops atlanticus) which contained information on body weight W as a function of fork length L and dorsal girth G for more than 1100 individuals from Florida, the Gulf of Mexico, Caribbean Sea and western Africa. A popular formula (W=G2L/800), developed more than a century ago specifically for tarpon was originally derived from geometric and physical relationships between fish body weight and shape. We found this model to be negatively biased >−15% (i.e., underestimates individual fish weights) across the tarpon's entire size range up to 130kg (the current world record). Bias for our “new” ellipsoid volumetric formulation, which extends the principles of allometry, was less than 1% across the entire range of observed sizes. Our new estimator was used by Texas Parks and Wildlife Department to assist fishery management in setting “minimum maximum” sizes for recreationally caught tarpon that have the potential to break the Texas State record. It has also supported catch-and-release tarpon fisheries and tournaments in Florida, Texas, Mexico, and Trinidad. It provides a reliable means to accurately estimate weights of tarpon of potential (world) record sizes.