Four Nordic temperature proxies based on tree growth at the northern timberline – ring-width from Sweden and Finland, maximum latewood density from Sweden, and height increment from Finland – were compared. Three indexing methods were used to enhance the low (centennial and above), medium (decadal-to-multidecadal) and high (decadal-to-interannual) frequencies. The proxies are shown to have a strong temperature signal (common variance) at the interannual-to-multidecadal scale, while the multidecadal-to-centennial trends are less coherent, perhaps reflecting intra-regional differences in growing conditions but more likely due to the more noisy regional curve standardization method used to retain the longest trends. Various methods of combining the four proxy series were explored and tested by comparison with four long temperature records from northern Fennoscandia. Only relatively high-frequency, spline-indexed series produced consistently positive verification statistics as a reconstruction model for summer temperature using all four proxies. Copyright © 2012 John Wiley & Sons, Ltd.