Greenshell™ mussel (GSM, Perna canaliculus) and king (Chinook) salmon (Oncorhynchus tshawytscha) are New Zealand's two major aquaculture species generating $380 million NZD in exports during the 2017–18 financial year. This study addresses the development and validation of a method based on Fourier transform—near infrared reflectance spectroscopy (FT‐NIRs) to determine proximate composition for both species to aid breeding‐, production‐ and consumer decisions. Rapid measurements of GSM (n = 176) were taken by FT‐NIRs and analysed by traditional wet chemistry ‘reference methods’ to develop calibration models for proximate composition (protein, moisture, fat, ash and carbohydrate). The predictive models for moisture (r2 = 0.98, root mean square error of cross validation (RMSECV) = 0.314, residual prediction deviation (RPD = 6.47), protein (r2 = 0.91, RMSECV = 0.295, RPD = 3.01)) and carbohydrate (r2 = 0.87, RMSECV = 0.440, RPD = 2.78) in GSM performed well. Additional models based on 90 portions of salmon were developed to predict moisture (r2 = 0.98, RMSECV = 1.02, RPD = 7), protein (r2 = 0.96, RMSECV = 0.347, RPD = 5.08), fat (r2 = 0.99, RMSECV = 1.09, RPD = 5.98) and ash (r2 = 0.72, RMSECV = 0.05, RPD = 1.9). The predictive FT‐NIRs and reference methods were tested for short‐term and intermediate precision, which demonstrated that the repeatability of the predictive models was comparable to the reference methods. Proximate analysis of GSM and king salmon using FT‐NIRs was quick (minutes for sample preparation and analysis rather than days) and all components were assessed simultaneously. This provides a low‐cost short turn‐around method suitable for industry and research applications.