Maximal or peak oxygen uptake (V˙O2 max and V˙O2 peak , respectively) are commonly measured during graded exercise tests (GXTs) to assess cardiorespiratory fitness (CRF), to prescribe exercise intensity and/or to evaluate the effects of training. However, direct measurement of CRF requires a GXT to volitional exhaustion, which may not always be well accepted by athletes or which should be avoided in some clinical populations. Consequently, numerous studies have proposed various sub-maximal exercise tests to predict V˙O2 max or V˙O2 peak . Because of the strong link between ratings of perceived exertion (RPE) and oxygen uptake (V˙O2), it has been proposed that the individual relationship between RPE and V˙O2 (RPE:V˙O2) can be used to predict V˙O2 max (or V˙O2 peak) from data measured during submaximal exercise tests. To predict V˙O2 max or V˙O2 peak from these linear regressions, two procedures may be identified: an estimation procedure or a production procedure. The estimation procedure is a passive process in which the individual is typically asked to rate how hard an exercise bout feels according to the RPE scale during each stage of a submaximal GXT. The production procedure is an active process in which the individual is asked to self-regulate and maintain an exercise intensity corresponding to a prescribed RPE. This procedure is referred to as a perceptually regulated exercise test (PRET). Recently, prediction of V˙O2max or V˙O2 peak from RPE:V˙O2 measured during both GXT and PRET has received growing interest. A number of studies have tested the validity, reliability and sensitivity of predicted V˙O2 max or V˙O2 peak from RPE:V˙O2 extrapolated to the theoretical V˙O2 max at RPE20 (or RPE19). This review summarizes studies that have used this predictive method during submaximal estimation or production procedures in various populations (i.e., sedentary individuals, athletes and pathological populations). The accuracy of the methods is discussed according to the RPE:V˙O2 range used to plot the linear regression (e.g., RPE9–13 versus RPE9–15 versus RPE9–17 during PRET), as well as the perceptual endpoint used for the extrapolation (i.e., RPE19 and RPE20). The V˙O2 max or V˙O2 peak predictions from RPE:V˙O2 are also compared with heart rate-related predictive methods. This review suggests that V˙O2 max (or V˙O2 peak ) may be predicted from RPE:V˙O2 extrapolated to the theoretical V˙O2 max (or V˙O2 peak) at RPE20 (or RPE19). However, it is generally preferable to (1) extrapolate RPE:V ˙ O 2 to RPE19 (rather than RPE20); (2) use wider RPE ranges (e.g. RPE ≤ 17 or RPE9–17) in order to increase the accuracy of the predictions; and (3) use RPE ≤ 15 or RPE9–15 in order to reduce the risk of cardiovascular complications in clinical populations.