As the inherent spatial variability of soil phosphorus (P) within a field is considerable, variable-rate (VR) technology of P fertilisation should be implemented over small areas (e.g. 1 m(2)). The objective of this study was to design and implement a soil sensor-based VR fertilisation system for on-the-go application of phosphate (P2O5) during maize planting. An on-the-go visible (VIS) and near-infrared (NIR) soil sensor with a measurement range of 305-1711 nm was installed at the front of a planter-applicator for on-the-go measurement of soil P. A previously developed VIS-NIR model was used to predict the extractable phosphorous (P-ext) and a custom-built LabVIEW programme was developed to record soil spectra, predict soil P-ext, calculate phosphate during on-the-go measurement and provide the signal to the fertiliser applicator to adjust the application rate. Alternate plots were used for VR application and for uniform-rate (UR) treatment. The number of plant leaves and grain yield were measured as growth indices that may be influenced by P deficiency. The coefficient of variation (CV) of P-ext measured on-the-go ranged from 5% to 51% while variation of phosphate ranged from 36% to 76% over the experimental plots. The average phosphate applied on VR plots was 28.75, 1.25 kg ha(-1) less than the UR (30 kg ha(-1)) recommended according to the standard soil test. The application rate of the phosphate ranged from 0 to 100 kg ha(-1) in the VR plots. Lower variation in plant leaves was observed in plots with VR treatment, possibly indicating better P distribution over the VR plots. The number of plant leaves variations were 25% and 31% for VR and UR plots, respectively However, there was no significant difference between VR and UR plots. The maize yield was significantly higher (336 kg ha(-1)) and less variable on plots that received VR treatment. (c) 2007 IAgrE. Published by Elsevier Ltd. All rights reserved.