Emerging international policy aimed at reducing carbon emissions from deforestation and forest degradation (REDD+) in developing countries, has resulted in numerous studies on above-ground live carbon (AGC) in tropical forests. However, few studies have addressed the relative importance of disturbance, topography, climate, soil and methods for stem measurement, on the estimation of AGC, or the costs of improving AGC estimates by altering sample regimes. We established 18 one hectare plots containing 7201 stems, stratified along forested elevation gradients in Tanzania. We recorded a broad set of physical, climatic and edaphic predictors of AGC and tree stature. AGC estimates using stem diameter, height and wood density, gave a mean value of 174.6 t ha−1, compared with 229.6 t ha−1 when height was excluded. Regression models revealed that stems were tallest for a given diameter at mid-elevation (1000–1250 m), on south-facing slopes, and without past logging. High AGC was strongly associated with shallow slopes, followed by intermediate elevation, elephant absence, low potential evapotranspiration and low soil pH. Further regression models to investigate the structural habitat features associated with AGC, revealed significant positive influence of basal area, stem density, and height:diameter ratio, rather than the mean wood density of species present. Large stems (⩾70 cm dbh; 4.6% of stems) contained 52% of AGC in all plots, declining to 36% in lowland plots. We discuss the cost:benefit of different measurements and recommend a tiered approach to AGC monitoring, depending on available resources. AGC assessments in African forests could exclude small stems, but should aim to record disturbance, topography and species. Stem height is vital for AGC estimation and valuation; when excluding height our 55 t ha−1 over-estimation of AGC would have over-valued the carbon resource by 24% (US$3300 ha−1).