Litter decomposition is frequently inferred from assays where mesh bags of leaf litter are incubated in the environment for a period of time and mass loss measured upon retrieval. Mass loss data are then overwhelmingly modeled assuming a negative exponential function whose central parameter, k, defines a constant rate of decay in litter breakdown. This constant decay assumption and k have been tremendously useful in ecosystem ecology, but obfuscates temporal patterns of non-constant decay. Although other models allowing time-varying decay exist, they are likely underutilized due to concerns about ease of implementation, translation of parameters across distinct models, greater data demands to identify parameters, and uncertainty about model choice. Here, I show that these concerns are addressed in part by several developments. First, current publicly available software enables relatively user-friendly fitting of a wide variety of non-linear models to litter mass loss data. Synthesized datasets help constrain fitted parameters, e.g., through use of Bayesian priors. Second, concerns about data sparsity and parameter translation across models and studies can be ameliorated by shifting focus from parameters to derived quantities, such as half-lives and instantaneous decay rates, which can be well-constrained even when individual parameters are poorly identified. Lastly, I introduce a new extension of existing two-pool models interpreted as capturing a rapidly leachable pool with a negative exponential function and the dominantly biological breakdown of another pool with a Weibull function. This model tightly captures smooth variation in synthesized time series. I then show with empirical data and data from simulations of process-based models how this non-linear modeling framework can lead to new insights about temporal trajectories (“shapes”) of mass loss, including methodological choices surrounding leaching, controls on mass loss such as temperature sensitivity, and uncertainty surrounding fragmentation rates. Flexible modeling of litter mass ontogenies is not always needed, but can accelerate quantitative understanding and prediction of leaf litter breakdown and stocks in diverse aquatic ecosystems.