Soil liquefaction is one of the major concerns causing damage to the structures in saturated sand deposits during earthquakes. Simplified methods for the assessment of liquefaction potential rely on the limit states that are generally established with built-in conservatism and a great deal of subjectivity. Well-known SPT- and CPT-based methods are widely used in the design practice for this purpose due to their simplicity and reasonable predictive capability. However, these methods do not account for various sources of uncertainties explicitly. Moreover, evaluations are made only at the locations of test results and are generalized for the whole region, which may not give accurate results where spatial variation of soil properties is significant. The present study focuses on the probabilistic evaluation of liquefaction potential of Alameda County site, California, considering spatial variation of soil indices related to CPT soundings. A stochastic soil model is adopted for this purpose using random field theory and principles of geostatistics by developing 2D exponential correlation functions. It has been observed that the probability of liquefaction is significantly underestimated as much as 34 %, if the spatial dependence of soil indices is not considered. Further, the effect of spatial variation is more prominent in low-level earthquakes compared to the high-level earthquakes, showing a 41.5 % deviation for magnitude 8.1 and a 60.5 % deviation for magnitude 5.0 earthquake at a depth of 10 m.