ABSTRACT We present a novel retrieval technique that attempts to model phase curve observations of exoplanets more realistically and reliably, which we call the 2.5-dimensional (2.5D) approach. In our 2.5D approach we retrieve the vertical temperature profile and mean gaseous abundance of a planet at all longitudes and latitudes simultaneously, assuming that the temperature or composition, x, at a particular longitude and latitude (Λ, Φ) is given by $x(Λ ,Φ) = \bar{x} + (x(Λ ,0) - \bar{x})\cos ^nΦ$, where $\bar{x}$ is the mean of the morning and evening terminator values of x(Λ, 0), and n is an assumed coefficient. We compare our new 2.5D scheme with the more traditional 1D approach, which assumes the same temperature profile and gaseous abundances at all points on the visible disc of a planet for each individual phase observation, using a set of synthetic phase curves generated from a GCM-based simulation. We find that our 2.5D model fits these data more realistically than the 1D approach, confining the hotter regions of the planet more closely to the dayside. We then apply both models to WASP-43b phase curve observations of HST/WFC3 and Spitzer/IRAC. We find that the dayside of WASP-43b is apparently much hotter than the nightside and show that this could be explained by the presence of a thick cloud on the nightside with a cloud top at pressure <0.2 bar. We further show that while the mole fraction of water vapour is reasonably well constrained to (1–10) × 10−4, the abundance of CO is very difficult to constrain with these data since it is degenerate with temperature and prone to possible systematic radiometric differences between the HST/WFC3 and Spitzer/IRAC observations. Hence, it is difficult to reliably constrain C/O.