Tiling array is a new platform designed to detect sequence variation as well as gene expression for novel alleles. Gresham et al. (2006) developed a program called SNPscanner for tiling array technology to detect single nucleotide polymorphisms between two strains. In this thesis, its performance is demonstrated and two improvements are proposed. The two main SNP detection problems are bias problem and the separation of two closely linked SNPs. For bias problem, it is not appropriate for SNPscanner to fit the model with symmetric assumptions since it is observed that when the SNP occurs close to the two ends of a probe, the decrease of the affinity is not symmetric. The model is corrected with asymmetric assumptions and it successfully solves the problem of detection bias. For the separation of two closely linked SNPs, the region of positive signals detected by SNPscanner sometimes includes more than one SNP. Curve fitting with Gaussian kernel is proposed and log-likelihood ratio test is performed to select the model with either one peak or two peaks. It gives the correct detection of existing SNPs and improves the accuracy of the detection. Although the proposed method in this study is on the right direction of improving the SNP detection with SNPscanner, there is still space to enhance the function of the software. For example, the log-likelihood ratio proposed in this study does not strictly follow the chi-square distribution and the proposed Gaussian kernel does not fit the curve well in certain circumstances. In addition to these, the model needs to record lots of information from the training data and limits the flexibility of detecting SNPs when applied on a different species. Further improvement is still under study.