The precast prestressed concrete pile (PPC) is one of the most common types of pile frequently adopted for large scale of building and hotel complex development in Macau. Once if pile foundation is designated, pile loading tests are necessary to be assigned to evaluate the performance of pile capacity; Static Load Test (SLT) is one of the acceptable tests to justify the actual resistance of pile on site. However, once SLT could not verify the pile capacity, further testing such as Pile Driving Analyzer (PDA) has to be proposed to figure out the conditions of the pile. Even though the capacities could be proved eventually, it is foreseen that additional time and cost for testing would increase significantly, and it is quite common to result in construction delay and over budget. A case happened in Cotai area in Macau, in which additional testing with PDA has to be proposed because of the failure of PPC piles in SLT, is adopted in this study, that attempted to find out the reason of failure in SLT and to investigate the possibility to predict such behavior effectively, such that engineer could be able to get preliminary and reliable predictions in SLT prior to the real test. Eventually prevention and remedial works could be made earlier to minimize the extra potential cost and time extensively. After careful review of all SLT test results, it is believed that the distribution of pile capacities between shaft friction and end-bearing is somehow related to the behavior of pile in SLT test. Thus, the evaluation of pile capacities is examined in this study accordingly. SPT-N values from site investigation are used to estimate the capacities of piles. An innovative procedure is proposed to interpolate the SPT-N values from the available boring logs at given locations. Firstly, the general regression neural network (GRNN) is utilized in the three-dimensional geological modeling with the given data. It is well known that soil spatial variation in different directions shall be considered in order to obtain reasonable results during numerical soil modeling. Therefore, the scale of fluctuation (SOF) is introduced in this study, to make optimized solution in GRNN modeling taking into account soil spatial variability. In general, GRNN would result in a relatively smooth profile against the observed data. However, because of the long preparation time for data reduction and optimization, this method is not feasible enough for practical engineer on site. Hence, an alternative method to interpolate SPT-N values is used. The principal component analysis (PCA) based on multivariate statistical analysis is also evaluated in this study. With relatively lesser time consumed in data optimization than the GRNN, PCA is beneficially suggested for site engineer to interpolate the SPT-N values. In general, PCA would result in a more variable profile compared with that from GRNN. By using both GRNN and PCA to interpolate the SPT-N values, the Meyerhof analytical and empirical methods are adopted in estimation of pile capacities. It is found that the overall prediction of pile capacities between these two SPT-N interpolation methods are more consistent in Meyerhof analytical method than in the empirical method. For this particular study, PCA with Meyerhof analytical is preferable to make reliable prediction in SLT beforehand on site.