Since the first successful Magnetic Resonance Imaging (MRI) image was produced by Paul Lauterbur in 1973, the field of MRI has been improving by leaps and bounds. The number of MRI and functional MRI (fMRI) papers have sky rocketed over the last decade, alongside with advancements in MRI field strength and techniques. In this thesis, I explore various methods for improving data quality for high resolution fMRI in 3T and 7T MRI scanners. Firstly, I studied the effect of Prospective Motion Correction (PMC) on 3T data using a simple visual paradigm. In contrast to most conventional techniques that use retrospective motion correction (RMC), PMC collects real-time motion data and uses it to update the acquisition field of view prior to each radiofrequency (RF) pulse. This allows for the correction of spin-history effects and intra-volume distortions. In this study, I utilized a secondary optical camera in the bore of the scanner to track a Moiré phase marker attached to the participant via a custom-moulded dental mouthpiece. I demonstrated that the camera is capable of accurately tracking the participant’s head motion. While simple metrics such as temporal signal-to-noise ratio (tSNR) and functional contrast-to-noise ratio (fCNR) showed no difference between the two methods, more complex analysis such as the Linear Discriminant Contrast (LDC) showed that the PMC data was indeed cleaner than the RMC data for higher resolution data. Next, I compared the sensitivity of two multi-voxel pattern analysis (MVPA) methods, Support Vector Machines (SVM) and Linear Discriminant Contrast (LDC). MVPA attempts to capture the relationship between the spatial fMRI activity and the experimental manipulations by treating it as a supervised learning problem. This is a promising technique that can capture spatial activation patterns that are lost in univariate analysis. I demonstrated through both actual fMRI data and computer simulations that LDC is a better MVPA metric than SVM. This agrees with our theory that SVM has more inherent variability and less sensitivity due to its limitations, discretization of results, rigid decision boundaries and ceiling effects. Subsequently, I analysed the quality of fMRI data acquired in a 3T Prisma scanner vs a 7T Terra scanner using a visual attention paradigm. While 7T scanners are becoming increasingly commonplace with over 70 of them worldwide now, the higher field strength also comes with its own host of problems. Field inhomogeneities and artefacts are a larger problem at 7T, and the smaller voxel sizes also cause data to be more susceptible to motion. As such, it is important to establish if there is a real benefit to using a 7T scanner. I observed that both 3T and 7T data showed similar trends with comparable z-scores and concluded that both scanners yielded comparable results. However, the 7T data was acquired at a much higher resolution (64x smaller volume per voxel) and thus, these results indicate a benefit of 7T as comparable results were achieved in spite of the smaller voxel volume. I hypothesized that acquiring data in a 7T scanner would be informative if studies sought to probe further into laminar or columnal structures which require submillimetre resolution, while a 3T scanner should suffice for studies looking at coarse regional activations. I did not explore the benefits of using 7T MRI at coarser resolutions. I also assessed the utility of boundary-based registration (BBR) realignment to improve on conventional RMC techniques to realign fMRI time series. Some motion artefacts affect the image in non-rigid ways and thus, voxel-based registration (VBR), generally utilized in conventional RMC, might be insufficient to properly realign fMRI time series. I demonstrated that BBR realignment outperforms VBR realignment across multiple metrics at submillimetre resolution, but no difference was observed at lower resolutions. Lastly, I examined the process of cleaning up 7T fMRI data for laminar analysis. Gradient echo (GE) sequences have been widely used for fMRI studies due to the high signal-to-noise ratio (SNR) and low specific absorption rate (SAR) relative to other sequences. However, GE sequences have been shown to exhibit superficial bias due to the presence of draining veins. I employed two methods- excluding venous voxels and utilizing a regression analysis, to remove superficial bias in an attempt to unmask any laminar effects for a visual attention task. In summary, I have explored various methods of optimizing fMRI data, ranging from initial setup decisions, such as which field strength scanner to use, to final MVPA analysis methods. I also analysed methods to remove motion artefacts, through both PMC and RMC, as well as post-processing methods to remove superficial bias in laminar data.