Soil salinity is best estimated by saturated paste extract (ECe), the most reliable monitoring method to assess plant growth that is directly related to the field water content. However, this procedure is laborious and time-consuming, therefore, more convenient methods such as 1:5 soil: water extract is commonly used to estimate the ECe. Traditionally, a conversion factor (CF) based on a linear correlation between the diluted extract and ECe is employed for the estimation purposes. However, CF is affected by site-specific conditions. The objective of this research is to demonstrate a novel modelling approach that allows incorporating site-specific soil and irrigation water parameters to improve the accuracy of the ECe estimation. A total of 177 soil samples were collected from agricultural soils in the Jordan Valley representing different soil textures, crops and water qualities. ECe, EC1:5, clay and sand content, soil texture and saturation percentage (θSP) were determined. The collected data were used to construct models using three distinct approaches: traditional CF; incorporating θSP as a surrogate of soil texture to cater the CF to the site-conditions, and Artificial Neural Networks to incorporate site-specific parameters. The neural network model gave the most accurate estimates (R2 = 0.987, MSE = 2.39) and was able to handle the heteroscedasticity of the data. Meanwhile, the incorporation of θSP to estimate the CF that best represent the site has shown improved prediction quality over the traditional CF approach as it was more capable of handling the heteroscedasticity of the data. The neural network model allows for the incorporation of location-specific parameters and therefore offers a flexible tool for better management of agricultural soils. Although, this work used a case study location to demonstrate the concepts discussed, the approach is generalizable and can be easily adapted to other locations.