Timely crop yield forecasts at regional and national level are crucial to manage trade and industry planning and to mitigate price speculations. Sugarcane is responsible for 70% of global sugar supplies, thus making yield forecasts essential to regulate the global commodity market. In this study, a sugarcane forecasting system was developed and successfully applied to São Paulo State, the largest cane producer in Brazil. The system is based on multiple linear regressions relating agro-climatic indicators and outputs of the sugarcane model Canegro to historical yield records. The resulting equations are then used to forecast the yield of the current season using 10-day period updated values of indicators and model outputs as the season progresses. We quantified the reliability of the forecasting system in different stages of the sugarcane cycle by performing cross-validations using the 2000-2013 time series of official stalk yields. Agro-climatic indicators alone explained from 38% of inter-annual yield variability (at State level) during the boom growth phase (i.e., January-April) to 73% during the second half of the harvesting period (i.e., September-October). When Canegro outputs were added to the regressor set, the variability explained increased to 63% for the boom growth phase and 90% after mid harvesting, with the best performances achieved while approaching the end of the harvesting window (i.e. at the beginning of October, SDEP = 0.8 t ha-1, R2cv = 0.93). It is concluded that the overall performances of the system are satisfactory, considering that it was the first attempt based on information exclusively retrieved from the literature. Further improvements to operationalize the system could be possibly achieved by the use of more accurate inputs possibly supplied by the collaboration with local authorities.