The simplest theory describing large-scale redshift-space distortions (RSD), based on linear theory and distant galaxies, depends on the growth of cosmological structure, suggesting that strong tests of general relativity can be constructed from galaxy surveys. As data sets become larger and the expected constraints more precise, the extent to which the RSD follow the simple theory needs to be assessed in order that we do not introduce systematic errors into the tests by introducing inaccurate simplifying assumptions. We study the impact of the sample geometry, non-linear processes and biases induced by our lack of understanding of the radial galaxy distribution on RSD measurements. Using Large Suite of Dark Matter Simulations of the Sloan Digital Sky Survey II (SDSS-II) luminous red galaxy data, these effects are shown to be important at the level of 20 per cent. Including them, we can accurately model the recovered clustering in these mock catalogues on scales 30–200 h−1 Mpc. Applying this analysis to robustly measure parameters describing the growth history of the Universe from the SDSS-II data gives f(z= 0.25)σ8(z= 0.25) = 0.3512 ± 0.0583 and f(z= 0.37)σ8(z= 0.37) = 0.4602 ± 0.0378 when no prior is imposed on the growth rate, and the background geometry is assumed to follow a Λ cold dark matter (ΛCDM) model with the Wilkinson Microwave Anisotropy Probe (WMAP)+Type Ia supernova priors. The standard WMAP constrained ΛCDM model with general relativity predicts f(z= 0.25)σ8(z= 0.25) = 0.4260 ± 0.0141 and f(z= 0.37)σ8(z= 0.37) = 0.4367 ± 0.0136, which is fully consistent with these measurements.