The transient response of a polymer-coated surface acoustic wave (SAW) vapour sensor depends on partitioning and diffusion of vapour species into the polymer in conjunction with its thickness and viscoelastic properties. The shapes of transient signals carry information about vapour identities due to specificity of the partition coefficient and the diffusion coefficient. The analysis of transient signals therefore offers a simpler approach for vapour identification in comparison to conventional electronic nose systems that employ a broadly selective sensor array. The transient response-based methods are however not developed to a similar level of maturity as their sensor array counterparts. The main reason for this is associated with complex signal generation kinetics and polymer viscoelasticity. The latter is independent of vapour identities (assuming low concentrations) but influences sensor response through nonlinear dependences on polymer thickness and viscoelastic coefficients. In this paper, we endeavour to find out whether viscoelasticity and its manifestation through thickness dependences could be turned into an advantage for transient-based vapour identification. Using an established SAW sensor model with additive noise we analyse sensor transients by wavelet decomposition and principal component analysis (PCA) for various combinations of polymer thickness, viscoelastic storage and loss moduli and noise level. We calculate vapour class separability measures defined on the basis of scatter matrices of principal components of wavelet coefficients to determine the discrimination ability of sensor transients for various combinations of film thickness and viscoelastic parameters. The simulation experiments are performed by considering a polyisobutylene-coated SAW oscillator sensor under exposure to seven volatile organic compounds (chloroform, chlorobenzene, o-dichlorobenzene, n-heptane, toluene, n-hexane and n-octane). The film thicknesses are varied from thin film (where mass loading dominates) to thick film through fundamental film resonance (where viscoelastic effects dominate) to very thick film regions spanning over a few higher order resonances. The storage and loss shear moduli are varied to simulate conditions of glassy, glassy-rubbery and rubbery phases of polymer. The transient response generation incorporates an additive noise source with uniform distribution over a specified range. Effect of noise variation on class separability is also studied. A comparison of the wavelet transform method is made with the phase space-based partial-least-squares regression method for feature extraction. In conclusion, it is found that (i) vapour class separability increases with polymer thickness for all viscoelastic conditions from glassy to rubbery, except near film resonances, (ii) near resonance class separability dramatically declines for films with no or low loss, (iii) by imparting finite viscoelastic losses to polymer coatings, not only are the detrimental effects of film resonance eliminated but also the sensor performance improves, and (iv) viscoelastic effects produce better noise immunity in thick film sensors. This analysis provides a new perspective for designing high-performance transient sensors through optimization of film thickness for specific polymer selections.