The circumstances that select for plant anti‐herbivore defences are not well understood. In New Zealand, the 'divaricate' cage‐like architecture of many woody plants may have arisen as a defence against avian browsing; it also has some ability to deter browsing by introduced deer. Its prominence on alluvial soils in frosty and droughty areas led us to hypothesize that structural defences are of most value where fertile soils coincide with climatic constraints that prevent plants from quickly growing out of the browse zone.We tested this hypothesis by growing seedlings of three divaricate species and their broadleaved congeners on fenced and unfenced plots beneath tree‐fall gaps at three humid frost‐free sites in the North Island of New Zealand, and at three colder and drier sites in the South Island. Soil total phosphorus levels were moderate to high at all sites (344–961 ppm), and pellet counts indicated similar average deer densities at North and South Island sites, enabling us to attribute variation in seedling growth and survival primarily to climatic differences between our sites.On fenced plots, average relative height growth rates in the North Island exceeded those on the southern sites by 64%, and growth of broadleaved species averaged 132% and 99% greater than that of divaricate congeners at North and South Island sites, respectively. On the northern sites, broadleaved species also held a large (137%) net height growth advantage over divaricates even on unfenced plots. In contrast, exposure to browsing on the coldest sites resulted in complete loss of the broadleaved net height growth advantage.Synthesis. Even in the presence of deer, divaricate plants were overgrown on humid frost‐free sites, but not on colder and drier sites, where browsing neutralized the superior growth potential of broadleaved species. The cost of structural defences therefore paid off only on sites where climatic adversity reduces the ability of plants lacking such defences to grow quickly out of the browse zone, in agreement with our hypothesis. [ABSTRACT FROM AUTHOR]