In this papel; a rational framework in classifying soil properties using systematic statistical approaches has been proposed and illustrated through a case study using some simple physical index properlies ofsoji marine clay in Peninsular Malaysia. The framework includes an initial data screening, follmved by trend analysis, removal ofoutliers, normality conformance and lastl), the descriptive statistics. Common erroneous treatments that used to be overlooked were examined in detail and the results ofanalyses were intelpreted slalislically incorporating engineeril1gjudgment. The effectiveness o/various engineering even among the researchers. The analytical tools commonly adopted in characterizing soi I properties are none other than some simple descriptive functions such as mean, range and simple regression analysis (e.g. Ting and Ooi, 1977; Kobayashi et a!., 1990; Mohamad et a!., 1994; Chen and Tan, 2003). Some of the reported values were potentially misleading as they were essentially extracted from raw data (soil samples are prone to various disturbances especially those from unsupervised slack site investigation works) without proper statistical treatment. The soil properties derived from a site investigation are always found widely scattered and limited in number. Consequently, the use of basic statistics to assess these highly variable natural materials may not be as directly applicable as in the case ofcontrollable manufactured materials. It should be noted that the statistical methods used to characterize the soil properties and hence determine its characteristic value have to be exercised with caution. Taking into account the comparable experience of those statistical methods, inappropriate selection of the characteristic value would either result in conservative design (over-design) which is uneconomical or causing various short and long-term serviceability and safety problems (under-design), of which both are undesirable. Lumb (1966) was among the first to show that most soil properties could be regarded as random variables conforming to the normal or Gaussian theoretical distribution. Consequently numerous statistical methods established based on the normal distribution may safely be applied in estimating design parameters. The statistical functions used in the study to check the normality of natural soil properties were chi-square test (c2-test) and the graphical plot of standardized normal variate. However, engineers may be puzzled if their soil data are found not normally distributed especially when the number of data is limited. Phoon and Kulhawy (1999) commented that most of the coefficient of variation (COY) reported in the literature were based on total variability analyses and therefore were considerably larger than the actual inherent soi I variability. The error can be attributed to four potential problems : (i) soil data from different geologic units are mixed, (ii) equipment and procedural controls generally are insufficient, (iii) deterministic trend in the soil data are not removed, and (iv) soil data are taken over a long time period. The same could be reasonably inferred from other reported statistical functions especially mean and standard deviation as they are all analysed at one time using similar approach. Orr (2000) discussed several statistical methods for determining the characteristic values in geotechnical designs in accordance to Eurocode common statistical methods were compared, contrasted and discussed. The proposed framework was found useful in characterizing soil properties and the important findings were listed.