Consider two different genes (call them two different loci, if you like). Gene A where + is a wild-type, functional copy of the gene, and a is a mutant, loss-of-function copy of the gene. Gene B where + is a wild-type functional copy of the gene and b is a mutant, loss-of-function copy of the gene.
An individual with this genotype:
+/+; +/+
Will be wild-type. And if the a allele of Gene A is recessive, and the b allele of Gene B is recessive, then an individual with this genotype:
+/a; +/b
Will also have a wild-type phenotype. In a diploid organism, dominance is when +/a individuals have a mutant phenotype. The same holds true for +/b individuals, if one copy of the mutant allele causes a mutant phenotype then the b allele is dominant to the + allele of the B gene.
If both mutant alleles are dominant then this individual would have two mutant phenotypes:
+/a; +/b
Most mutations, however, are recessive to the wild-type allele, so you would not see a mutant phenotype until there are two mutant copies and no wild-type alleles. Either:
a/a; +/+
or:
+/+; b/b
The double mutant would have this genotype:
a/a; b/b
You cannot interpret, or make logical sense, of an epistasis experiment unless the alleles are recessive.
If this individual does not have a double mutant phenotype, but instead only shows the 'A' gene loss-of-function phenotype:
a/a; b/b
Then we would say that the A gene is epistatic to the B gene, because homozygous A gene mutants "hide" the B gene mutant phenotype.
If, instead the double mutant only has the 'B' gene mutant phenotype then we would say that the B gene is epistatic to the A gene.
If the double mutant has both phenotypes then there is no epistasis between these two genes, or loci. Conversely, if the double mutant has a wild-type phenotype then the mutations are suppressing each other.