The work aims to explore a novel detection approach to realize the high-sensitivity detection of toxicants in foods like fruit and vegetable. The multi-branched nano-gold with high SERS activity was taken as SERS substrate, and substrates loaded with phoxim, chlorpyrifos and bisphenol A were subjected to Raman detection to collect the standard spectra of different detection objects for qualitative analysis. Quantitative detection models of different substances were constructed. The above approach was introduced to the toxicant qualitative detection of actual food samples for practical experiment. The DERS spectra of three substances for qualitative detection were successfully collected through experiment and the detection limit of above substances by the approach was 4.97, 0.565 and 0.594 ng/L respectively. When the above approach was used in the detection of actual sample, the recovery rate of phoxim and chlorpyrifos in fruit and vegetable and the residues of bisphenol A in drinks with plastic package was over 80%. The research result provides a theoretical basis for reducing the detection limit of phoxim, chlorpyrifos and bisphenol A in foods like fruit and vegetable.