Extensive degradation of ecosystems, combined with the increasing demands placed on the goods and services they provide, is a major driver of biodiversity loss on a global scale. In particular, the severe degradation of large rivers, their catchments, floodplains and lower estuarine reaches has been ongoing for many centuries, and the consequences are evident across Europe. River restoration is a relatively recent tool that has been brought to bear in attempts to reverse the effects of habitat simplification and ecosystem degradation, with a surge of projects undertaken in the 1990s in Europe and elsewhere, mainly North America. Here, we focus on restoration of the physical properties (e.g. substrate composition, bank and bed structure) of river ecosystems to ascertain what has, and what has not, been learned over the last 20 years. First, we focus on three common types of restoration measures—riparian buffer management, instream mesohabitat enhancement and the removal of weirs and small dams—to provide a structured overview of the literature. We distinguish between abiotic effects of restoration (e.g. increasing habitat diversity) and biological recovery (e.g. responses of algae, macrophytes, macroinvertebrates and fishes). We then addressed four major questions: (i) Which organisms show clear recovery after restoration? (ii) Is there evidence for qualitative linkages between restoration and recovery? (iii) What is the timescale of recovery? and (iv) What are the reasons, if restoration fails? Overall, riparian buffer zones reduced fine sediment entry, and nutrient and pesticide inflows, and positive effects on stream organisms were evident. Buffer width and length were key: 5–30 m width and > 1 km length were most effective. The introduction of large woody debris, boulders and gravel were the most commonly used restoration measures, but the potential positive effects of such local habitat enhancement schemes were often likely to be swamped by larger-scale geomorphological and physico-chemical effects. Studies demonstrating long-term biological recovery due to habitat enhancement were notable by their absence. In contrast, weir removal can have clear beneficial effects, although biological recovery might lag behind for several years, as huge amounts of fine sediment may have accumulated upstream of the former barrier. Three Danish restoration schemes are provided as focal case studies to supplement the literature review and largely supported our findings. While the large-scale re-meandering and re-establishment of water levels at River Skjern resulted in significant recovery of riverine biota, habitat enhancement schemes at smaller-scales in other rivers were largely ineffective and failed to show long-term recovery. The general lack of knowledge derived from integrated, well-designed and long-term restoration schemes is striking, and we present a conceptual framework to help address this problem. The framework was applied to the three restoration types included in our study and highlights recurrent cause–effect chains, that is, commonly observed relationships of restoration measures (cause) and their effects on abiotic and biotic conditions (effect). Such conceptual models can provide useful new tools for devising more effective river restoration, and for identifying avenues for future research in restoration ecology in general.