The three-dimensional (3D) Look–Locker (LL) acquisition is a widely used fast and efficient T1 mapping method. However, the multi-shot approach of 3D LL acquisition can introduce reconstruction artifacts that result in intensity distortions. Traditional 3D LL acquisition generally utilizes a centric encoding scheme that is limited to a single phase-encoding direction in k space. To optimize k-space segmentation, an elliptical scheme with two phase-encoding directions is implemented for the LL acquisition. This elliptical segmentation can reduce the intensity errors in the reconstructed images and improve the final T1 estimation. One of the major sources of error in LL-based T1 estimation is a lack of accurate knowledge of the actual flip angle. A multi-parameter curve-fitting procedure can account for some of the variability in the flip angle. However, curve fitting can also introduce errors in the estimated flip angle that can result in incorrect T1 values. A filtering procedure based on goodness of fit (GOF) is proposed to reduce the effect of false flip angle estimates. Filtering based on GOF weighting can remove probable incorrect angles that result in bad curve fitting. Simulation, phantom and in vivo studies have demonstrated that these techniques can improve the accuracy of 3D LL T1 estimation. Copyright © 2013 John Wiley & Sons, Ltd.