Solenoid proteins are proteins with repeats of 5 to 40 residues in length. Identifying solenoid proteins presents a big challenge because the repeat sequences are highly degenerated. Here, we present a nearest neighbor (NN) method for predicting solenoid proteins based on residue composition. The distance between proteins is calculated as a weighted Euclidean distance defined by the residue composition vector. The NN method predicts solenoid proteins with an overall accuracy of 95.5% with 94.3% sensitivity and 96% specificity, outperforming other methods in direct comparisons. We also demonstrate that combining the NN method with HHrepID and Trust, which are previously published methods for addressing the same problem, can dramatically reduce the false positive rates in predicting repeats.