If you know the number of modes in the data - k, apriori, then why not use it to cluster the data into k clusters using Kmeans function.
Once you have the clusters, finding the mean and std-dev of each cluster should lead you two gaussian distributions.