Purpose Recent advances in stochastic frontier analysis (SFA) suggest two alternative approaches to account for unobserved heterogeneity and to distinguish between persistent and transient inefficiency. The first approach is the generalized true random effects (GTRE) model, and the second approach is an autoregressive inefficiency (ARI) model. This study compares them to highlight whether they capture similar inefficiency aspects. Design/methodology/approach Using recent methodological advances in SFA, the authors estimate the GTRE and the ARI models using a Monte Carlo experiment and two real datasets from two industries (banking and agriculture). Findings The authors find that the two models provide quite different results in terms of inefficiency persistence and overall inefficiency (combination of transient and persistent inefficiency), regardless of the dataset considered. Practical implications The study findings suggest that researchers should be careful when referring to these two models because they do not capture the same inefficiency aspects, even though they have the same conceptual basis. This work is a warning about the empirical aspects of the persistent and transient efficiency framework, in order to convey a consistent story to the reader on firms' performance. Originality/value Even though they are used in a large number of studies, the present paper contributes to the productivity and efficiency literature by providing the first comparison of the GTRE and the ARI models.