Precision Agriculture (PA) and Remote Sensing (RS) technologies are increasingly being used as tools to assess crop and soil properties by breeders and physiologists.  These technologies are showing potential to improve genotype selections over their traditional field measurements, by providing quick access to crop properties throughout the crop cycle and yield estimation. The objective of this work was to use vegetation indices (VIs) and soil apparent electrical conductivity (EC_a) as predictor variables of yield. This information was obtained from a durum wheat yield trial, aiming to estimate yield of different genotypes under full and reduced irrigation. This work was carried out at CIMMYT’s experiment station at Ciudad Obregón/Sonora, Mexico, during 2013 wheat crop cycle.  There were four yield trials, two with reduced irrigation (RID) and two with full irrigation (FIG), which tested 112 different genotypes in a completely randomized design with three replications. A flight campaign took place, with six flights, once per week from March to April 2013, using a 6-channel multispectral camera with 10 nm FWHM filters onboard an airplane flying 300 m above ground yielding 0.3 m resolution. The EC_a data was collected just before sowing using an EM38 device in each plot. Twenty three different VIs ranging from chlorophyll, structural, red edge ratios and RGB indices were calculated using the multispectral images. A Pearson’s correlation was done using the yield of the check genotypes of each experiment with the VIs of each image and EC_a, aiming to explore the potential of each variable on predicting yield. This approach was followed by a subset multiple regression method, using as predictive variables the VIs coefficients fitted to each genotype considering a quadratic effect plus EC_a, to fit the yield of each genotype in a training dataset, and then applied into a Bootstrap method in the cross validation dataset. The significant correlations among yield from the check genotypes and VIs from all images, plus EC_a, ranged from -0.82 to 0.73 in the RID and from -0.70 to 0.60 in the FIG experiment. The correlation coefficients between measured versus predicted yield by the models got mean values of 0.51 (RID) and 0.68 (FIG) using the cross validation dataset, being 0.27 (RID) and 0.47 (FIG) of r-squared, indicating that the use of different VIs together may improve the yield prediction of breeding experiments.