I've used both on the same input files, with essentially default parameters. I am using multiple samples that are supposedly from two populations.
freebayes --fasta-reference reference.fa reads.bam > out.vcf
samtools mpileup -uf ref.fa aln1.bam aln2.bam | bcftools view -bvcg - > var.raw.bcf
bcftools view var.raw.bcf | bcftools view var.raw.bcf > var.flt.vcf
Samtools identifies snps that freebayes does not, and its very strange. For example, one very strong snp (quality of 999, looks good in tview in all samples) does not show up in freebayes. What is weird is that it does show up if you only process one population into vcd, the population that is almost entirely homozygous reference. The second population, which is fixed homozygous alternate, if included in the analysis means the SNP doesn't show up. Which is really weird.
I have no idea what is different about these snp calling methods such that something like this would happen, it is not a snp that there is much doubt about. Does anyone know?