Statistical methods such as survival analysis (parametric and non-parametric) and logistic regression, along with other non-parametric methods such as Consistent Threshold Estimate and Certainty method are used for generating injury risk curves from biomechanical data. Recently, much attention has been drawn to the question of which statistical methodology is more appropriate in the construction of risk curves for biomechanical datasets. Most of the papers and reports focus on existing biomechanical datasets for which they generate various risk curves using parametric and non-parametric methods and then suggest the use of one method over another based on some sort of criteria. The purpose of this paper is to look at the same statistical methods, but from the “inverse perspective”, e.g. evaluate different statistical methods using non-correlated, randomly generated data and to see if any of the widely used methods would yield a “good” risk curve when they are supposed to yield a “bad” risk curve. The “goodness” of a risk curve was evaluated based on 95% confidence intervals, the shape of the curve, and “goodness of fit” statistics. If the risk curve had a well pronounced S-shape, narrow confidence intervals and good “goodness of fit” statistics, then the method was concluded to be inappropriate for non-correlated datasets as it was expected to yield poor S-shape, wide confidence intervals and poor “goodness of fit” statistics. A well-correlated, randomly generated dataset was also evaluated using the various statistical methods. It was observed that logistic regression was able to clearly identify both the non-correlated and well-correlated datasets but suffered because of the underlying distribution that sometimes resulted in non-zero injury probability at zero stimulus level. Survival analysis with different types of censoring and underlying distributions was closely studied. Survival Analysis with a Weibull/ Log-Logistic/ Log-Normal underlying distribution and left- right censored data was not only able to clearly identify both non-correlated and wellcorrelated datasets, but also gave zero injury probability at zero stimulus level. This paper presents a new perspective of judging the applicability of the various statistical methods and recommends the statistical method, censoring technique, and the distributions that may be used for generating injury risk curves from biomechanical datasets