The Behavioral Portfolio Theory (BPT) developed by Shefrin and Statman is often confronted to the Markowitz's Mean Variance Theory (MVT). Although the BPT optimal portfolio is theoretically not mean variance efficient, some recent studies show that under the assumption of normally distributed returns, MVT and some models incorporating features of BPT can generate similar asset allocations. In this paper, we compare the asset allocations generated by BPT and MVT without restrictions. Using US stock prices from the CRSP database for the 1995-2011 period, we empirically determine the BPT optimal portfolio. We show that the Shefrin and Statman's optimal portfolio is MV efficient in more than 70% of cases. However, our results also indicates that MV investors will typically not select the BPT portfolio as this portfolio is always associated with a high return and an important level of risk. We show that the risk aversion coefficient of the BPT portfolio is up to 60 times smaller than the risk aversion degree of usual MV investors.