length-at-age (LAA) information acts as an essential part of integrated stock assessments, including directly impacting estimates of productivity. LAA data is often available at multiple stages in a fishery’s history and can be specified as such in most modern (integrated) assessment models. Specifying LAA data as a time series acknowledges that LAA changes in a non-random manner over time and is therefore non-stationary (i.e., varies over time). Non-stationary LAA data may reflect important fluctuations in stock productivity that if left undetected could result in biases in the stock assessment that may have implications for the sustainable management of the stock. LAA is known to change over time due to non-linear density-dependent (e.g., recruitment and fishing pressure) and independent (e.g., water temperature) drivers for several fish species. Arguably, LAA data should only be specified as non-stationary if sufficient evidence exists that temporal variation in LAA is non-random, with or without identifying the driver(s). Such evidence was obtained in 2007 for eastern Tasmanian banded morwong (Cheilodactylus spectabilis) supported by a general linear model describing non-random increases in mean length at several important age classes over time. LAA data has since been specified as non-stationary in the assessment model. However, it is unclear whether this specification remains appropriate since the observed non-stationarity in LAA has not subsequently been re-evaluated considering data available since 2007. The additional data may not have maintained the increase over time implied by the general linear model and may instead have fluctuated in a non-linear way. This study investigates the current evidence for non-random temporal variation in LAA of eastern Tasmanian banded morwong provided the updated data set. The general linear model approach and two non-linear approaches (generalized additive model and Gaussian process model) are applied to evaluate their relative utility for identifying non-random variation of mean length at key age classes over time. Contrary to the analysis in 2007, the general linear model now suggests LAA variation is random over time. Both non-linear models suggested otherwise and fit the data better according to residual plots, suggesting LAA data should continue to be specified as non-stationary in the stock assessment. Overall, this study found that non-linear models are more appropriate for evaluating the stationarity of LAA and advises which non-linear model may be more advantageous depending on the situation. This method can be applied to any fishery where an evidence-based decision needs to be made whether to specify LAA as a stationary process or not within an assessment model.