This paper develops a semi-parametric model for estimating ƒ(0), the probability density function at perpendicular distance zero from data obtained by the line transect method. A nonparametric kernel method produces asymptotic unbiased estimator for ƒ(0) provided that the detection model has a shoulder at distance zero, that is, = 0. In this paper we investigate a semi-parametric method for combining the negative exponential model-which does not have a shoulder at the origin-and the kernel model to fit the line transect data. Under this model a new estimator for ƒ(0) which is based on testing the shoulder condition to estimate the weight parameter, m between the combined models is developed. The properties of the new esimtator is addition to other estimators are studied for a wide variety of underlying models by using simulation techniques. The results demonstrate the superiority of the new estimator in most cases considered.