Published information is lacking on whethergenomewide selection, based on a single testerand a single year of testing, can identify maize(Zea mays L.) lines that would perform well inmultiple subsequent years and with multipletesters. Our objectives were to determine (i) ifphenotypic selection or genomewide selectionis more predictive of maize performancein future environments and with different testers,(ii) if combining both marker and phenotypicinformation is advantageous in selection,and (iii) the upward bias in correlations betweenmarker-predicted values and phenotypic values(rMP) when cross-validation across individualsand across environments is not performed. Weevaluated four elite populations, each with 150or 250 doubled haploid (DH) lines, in 18 environmentsin the US Corn Belt. The DH lines weregenotyped with 3072 single nucleotide polymorphismmarkers. The rMP values were 0.14 to 0.66for grain yield and 0.49 to 0.66 for moisture. Phenotypicselection was always as effective as ormore effective than genomewide selection. The rMP was lower when different testers were usedfor the training and test populations. Selectionbased on marker and phenotypic informationwas slightly more effective than genomewideselection or phenotypic selection alone. Thecorrelation between predicted and observedperformance was higher when cross-validationacross individuals and years was not performedthan when cross-validation across individualsand years was performed. For genomewideselection to be superior to phenotypic selection,the gains must not be measured in terms of theper-generation response with equal populationsizes and selection intensities.