The majority of our insight about glomerulonephritis (GN) is from observational research. Because proteinuria is an important element of outcome in GN, the validity of observational analyses is dependent on the metric used to model proteinuria. Previous metrics of proteinuria included the value at baseline, the average of all values over the entire follow-up (time-averaged), the instantaneous value at each time point (time-varying), or the average of all values prior to each time point, and each of these standardized to body surface area. It was not known which of these metrics best accounts for the risk of renal outcome and should be used in GN research. To address this, we studied 1351 adult patients with IgA nephropathy, focal segmental glomerulosclerosis, and membranous nephropathy from the Toronto GN Registry. Cox regression models for the risk of end-stage renal disease or a halving of estimated glomerular filtration rate included each proteinuria metric and were compared using model fit and discrimination. Proteinuria did not need to be standardized to body surface area. Time-varying proteinuria was the best metric to account for the prognostic effects of proteinuria over time, especially in focal segmental glomerulosclerosis and IgA nephropathy over the majority of follow-up, and in membranous nephropathy earlier in the disease course. Using alternate proteinuria metrics biased analyses up to 30.3%. These findings can improve the validity and design of future observational and prediction modeling studies in GN.