Lemon myrtle (LM), Backhousia citriodora, is a popular flavouring agent and herbal tea from Australia. To ensure traceability and consumer trust in global food supply chain, rapid and non-destructive tools are crucial. In this study, hyperspectral images were acquired from 91 L M samples sourced from three different origins (Malaysia, Queensland, and New South Wales (Australia)), within 950–2500 nm range. Classification models were developed using linear partial least squares-discriminant analysis (PLS-DA) with two approaches, pixel-based (trained by all spectral data points) and sample-based (trained by average spectrum). All models achieved classification accuracies above 96%. The sample-based PLS-DA model, trained by mean-centring transformed data, demonstrated the highest discriminatory performance. Both approaches show potential for LM origin classification, but the sample-based model is more suitable for automated and rapid industry applications due to its shorter calculation time. However, additional spectral data acquisition is necessary to improve the model and fully explore its capabilities and limitations.