Energy landscape theory is a powerful tool for understanding the structure and dynamics of complex molecular systems, in particular biological macromolecules. The primary sequence of a protein defines its free-energy landscape and thus determines the folding pathway and the rate constants of folding and unfolding, as well as the protein's native structure. Theory has shown that roughness in the energy landscape will lead to slower folding, but derivation of detailed experimental descriptions of this landscape is challenging. Simple folding models show that folding is significantly influenced by chain entropy; proteins in which the contacts are local fold quickly, owing to the low entropy cost of forming stabilizing, native contacts during folding. For some protein families, stability is also a determinant of folding rate constants. Where these simple metrics fail to predict folding behaviour, it is probable that there are features in the energy landscape that are unusual. Such general observations cannot explain the folding behaviour of the R15, R16 and R17 domains of alpha-spectrin. R15 folds approximately 3,000 times faster than its homologues, although they have similar structures, stabilities and, as far as can be determined, transition-state stabilities. Here we show that landscape roughness (internal friction) is responsible for the slower folding and unfolding of R16 and R17. We use chimaeric domains to demonstrate that this internal friction is a property of the core, and suggest that frustration in the landscape of the slow-folding spectrin domains may be due to misdocking of the long helices during folding. Theoretical studies have suggested that rugged landscapes will result in slower folding; here we show experimentally that such a phenomenon directly influences the folding kinetics of a 'normal' protein, that is, one with a significant energy barrier that folds on a relatively slow, millisecond-second, timescale.