Abstract The anthropogenic abundance of reactive nitrogen (N) forms has increased in the last few decades, increasing food production, but also resulting in increased eutrophication, algae blooms, loss of biodiversity, and greenhouse gas (GHG) emissions, in aquatic and terrestrial ecosystems. Denitrification beds are one approach to return this reactive N back to the atmosphere. These beds are large containers filled with a carbon (C) substrate, often wood byproducts. This substrate acts as a C and energy source for denitrifiers to reduce nitrate (NO₃⁻) from point source discharges into non-reactive dinitrogen (N₂) gas. This study investigated the biological mechanisms, controlling factors and adverse effects of NO₃⁻ removal in a woodchip denitrification bed (176 m x 5 m x 1.5 m) treating glasshouse effluent, and in barrels (0.2 m³) testing alternative carbon substrates for use in denitrification beds (pine and eucalyptus woodchips, sawdust, green waste, maize cobs and wheat straw). Furthermore, different techniques for measuring denitrification rates were compared and an approach for determining reliable NO₃⁻ removal rates in denitrification beds was developed. The NO₃⁻-N removal rates of the large denitrification bed averaged 7.6 g N m⁻³ bed volume d⁻¹ and increased with increasing temperature (Q₁₀ = 2.1). Microbial denitrification was the main NO₃⁻ removal mechanism in the denitrification bed and was always limited by C, rather than by NO₃⁻ availability. Dissimilatory nitrate reduction to ammonium (DNRA) and anammox were likely minor processes due to low ammonia (NH₄⁺) and nitrite (NO₂⁻) concentrations throughout the bed. Sulfate (SO₄²⁻) reduction, and methanogenesis, could not compete with NO₃⁻ reduction for C due to continuously high NO₃⁻ concentrations in the bed (>37 mg N L⁻¹). Aerobic processes dominated in the first few meters of the bed and close to the surface, but dissolved oxygen (DO) concentrations decreased rapidly along the bed from the inlet and remained low throughout most of the bed. There were some adverse effects observed in the denitrification bed associated with NO₃⁻ removal. About 4.3% of NO₃⁻-N removed from the bed was released as nitrous oxide (N₂O), but methane (CH₄) emissions from the surface of the bed were very low. A total of 35.4 kg d⁻¹ of carbon dioxide (CO₂) was released from the bed, but was not considered to contribute to a net increase in CO₂ concentrations of the atmosphere as the substrate (woodchips) used in the bed would likely decayed to CO₂ if used for other purposes. A net dissolved organic carbon (DOC) loss from the outlet was not detected. Longevity of the C substrate of the denitrification bed to support denitrification was about 39 years as calculated from the total C losses (CO₂ emissions and release of dissolved CO₂ and DOC from the bed). In a barrel study of different carbon substrates, NO₃⁻ removal was predominantly limited by C availability and temperature (Q₁₀ = 1.2) when NO₃⁻-N concentrations were above 1 mg L⁻¹. All C substrates showed high numbers of denitrification genes (nitrite reductase, nirS and nirK; nitrous oxide reductase, nosZ), providing further support that microbial denitrification was responsible for NO₃⁻ removal. Substrates incubated at 27.1 °C had greater ratio of nir/nosZ genes than substrates incubated at 16.8 °C, which was possibly a partial explanation for higher N₂O production in the warmer barrels. Wheat straw released 10% of NO₃⁻-N removed as dissolved N₂O, while all other carbon substrates released on average about 1.4% of the removed NO₃⁻-N as dissolved N₂O. Methane production occurred when NO₃⁻ concentrations were below 2 mg L-1 in the barrels. Maize cobs removed about 2.5 times more NO₃⁻ than woodchips, but released total organic carbon (TOC) in the outflow and a substantial portion of C was likely consumed by non-denitrifiers. Woodchips had low adverse effects and provided ideal conditions for denitrifiers determined by the relatively high ratio of denitrification gene copies/16S rRNA copies compared to the other C substrates examined. Investigating different approaches to determine denitrification rates revealed that both the acetylene inhibition method and the copy number of nitrite reductase genes (nirS, nirK) were useful for comparative estimations of NO₃⁻ removal rates between different carbon substrates and temperatures. However, neither approach could be used to quantify actual rates of denitrification. The acetylene inhibition method overestimated the actual NO₃⁻ removal rate by five fold. An in situ push-pull test using enriched ¹⁵NO₃⁻ was useful for determining denitrification rates at one specific point in a denitrification bed but would require multiple testing sites to obtain an average rate of NO₃⁻ removal for the bed. Comparing the ratio of the slopes of natural abundance ¹⁵N-N₂ and ¹⁵N- NO₃⁻ along the length of the bed determined the portion of NO₃⁻ removed by microbial denitrification, but not the denitrification rate. Measurements of dissolved N₂ concentration along the length of the bed were a useful approach to determine denitrification rates. This last approach was rapid and produced relatively accurate rates of NO₃⁻ removal compared to the other approaches conducted in this study. In summary, denitrification beds are an efficient approach for removing NO₃⁻ from point source discharges, but the beds do produce some N₂O. Woodchips could be combined with maize cobs to enhance NO₃⁻ removal rates while keeping adverse effects low in denitrification beds. Measurement of N₂ concentrations along the length and water flow of the bed was the most appropriate approach to determine denitrification rates of denitrifying bioreactors, and may also be useful in other ecosystems with high NO₃⁻ concentration and even flow.