• We conducted a field study to assess potential risks posed by C. urabae to the non-target N. annulata. • Sentinel N. annulata larvae and releases of C. urabae were conducted in habitats either with or without the presence of the target host U. lugens and its host plant. • In the overlapping choice habitat (i.e., both target and non-target hosts present), C. urabae was recovered from 3.1% of N. annulata. • In the habitat devoid of its target larvae, C. urabae parasitised only 0.2% of N. annulata. • Even though N. annulata was at greater risk of attack where the habitat overlap was most significant, it is unlikely to form a self-sustaining population upon this non-target species. • Generalist parasitoids Diolcogaster perniciosus and Meteorus pulchricornis had a much greater impact upon N. annulata than C. urabae. Cotesia urabae is biological control agent (BCA) that was introduced in New Zealand in 2011 against the Eucalyptus defoliating lepidopteran pest Uraba lugens. During pre- and post-release host specificity testing, one non-target moth, the New Zealand endemic Nyctemera annulata, was occasionally attacked, albeit a poor physiological host for C. urabae development. With the BCA now established in the New Zealand environment, we undertook a sentinel larval trial in the field, which included artificial releases of C. urabae, to assess non-target impacts in habitats, either with or without the presence of the target host and its host plant. Approximately two-thirds of all sentinel larvae released in the field were successfully recovered and reared in the laboratory to obtain parasitism. In the overlapping choice habitat (i.e., both target and non-target hosts present along with their host plants), the BCA C. urabae was recovered from 22.3% and 3.1% of target and non-target larvae, respectively. In the habitat devoid of Eucalyptus and its target larvae, C. urabae parasitised only 0.2% of the non-target, suggesting the BCA had immediately dispersed from release sites devoid of host cues. The non-target N. annulata was at greater risk of attack where the habitat overlap was most significant. This provided invaluable data to calibrate non-target risk models. Most importantly, the generalist parasitoids Diolcogaster perniciosus and Meteorus pulchricornis (Braconidae), which were also recovered in our field trials, had a much greater impact upon N. annulata than the specialist BCA. [ABSTRACT FROM AUTHOR]