There have been many studies about marginal discrepancy of single restorations made by various systems and materials. But many of statistical inferences are not definite because of sample size, measurement number, measuring instruments. etc. The purpose of this study was to compare the marginal adaptations of the anterior single restorations made by different systems and to consider more desirable statistical methods in analysing the marginal fit. The in vitro marginal discrepancies of three different all-ceramic crown systems (Celay In-Ceram. Conventional In-Ceram. IPS Empress 2 layering technique) and one control group (PFM) were evaluated and compared. The crowns were made from one extracted maxillary central incisor prepared with a 1mm shoulder margin and $6^{circ}$ taper walls by milling machine. 10 crowns per each system were fabricated. Measurements or a crown were recorded at 50 points that were randomly selected for marginal gap evaluation. Non-parametric statistical analysis was performed for the results. Within the limits of this study, the following conclusions were drawn: 1 Mean gap dimensions and standard deviations at the marginal opening for the maxillary incisor crowns were $98.2{pm}40.6{mu}m$ for PFM, $83.5{pm}18.7{mu}m$ for Celay In-Ceram, $104.9{pm}44.1{mu}m$ for conventional In-Ceram, and $45.5{pm}11.5{mu}m$ for IPS Empress 2 layering technique. The IPS Empress 2 system showed the smallest marginal gap (P<0.05). The marginal openings of the other three groups were not significantly different (P<0.05). 2 The marginal discrepancies found in this study were all within clinically acceptable standards ($100sim150{mu}m$). 3. When the variable is so controlled that the system may be the only one, mean value is interpreted to be the marginal discrepancy of a restoration which is made by each system and standard deviation is to be technique-sensitivity of each one. 4. From the standard deviations. the copy-milling technique (Celay/In-Ceram) was not considered to be technique-sensitive in comparison with other methods. 5. Parametric analysis is more reliable than non-parametric one in interpretation of the mean and standard deviation. The sample size of each group has to be more than 30 to use parametric statistics. The level of clinically acceptable marginal fit has not been established. Further studies are needed.