Cocoa is among the most desirable and palatable food in the world. Soil under cacao trees in Papua New Guinea (PNG) is fertile, however, it's been identified that to sustainably reach a production goal of 310,000¿t by 2030, PNG needs soil-based recommendations. For this purpose, a cost-effective, rapid and non-invasive soil testing approach using near infrared diffuse reflectance spectroscopy (NIR-DRS), to cater for soil testing needs of numerous small landholdings was assessed. This study investigated whether spectral data collected from smallholder farms across provinces can be used to estimate soil conditions for cacao farming. To assess this, a spectral library was generated by measuring 17 different soil parameters and spectral reflectance data over visible to near-infrared region (350–2500¿nm) in 507 soil samples collected from four production systems in Autonomous region of Bougainville (AroB), New Ireland province (NIP), East New Britain province (ENB) and East Sepik province (ESP). Analysis of spectral data showed that the shallow limestone soil of NIP is dominantly montmorillonite mixed with kaolinite clay minerals, while the three other sampling sites are mostly vermiculite with mixed illite type of clay. Among Cubist, partial-least-square regression (PLSR), and support vector regression (SVR) chemometric approaches, Cubist model outperformed both SVR and PLSR. Most soil properties were estimated with high accuracy (concordance coefficient¿=¿0.7) using the Cubist model, except for pH, electrical conductivity (EC), available K and B. The high quality of the prediction models maybe owing to the variability in properties of the four different soils captured through conditioned Latin hypercube sampling. Results from this study also suggest that soil samples collected from multiple depths may be used for predicting both surface (0–30¿cm) and sub-surface (30–90¿cm) soil properties. Thus, this tool can help reduce the cost of laboratory analysis, even if it requ