you need to give more details in order to receive more targeted advices.
apart from quality (specificity) of the primers and of the samples (contaminations?), maybe increasing somewhat (1-2°c) the annealing temperature could give you more reliable results. addition of molecules as DMSO or betaine (no more than 5% of total volume) could also help you.
I agree with Frederic Lepretre but until we get more detail 2 possibilities are alternatively spliced variants and contaminating genomic dna producing non specific bands. While you are trying Fred,s suggestion you could also pcr the primers on genomic dna and see if it produces the unexpected bands, If the bands are quite close togwther you might have a polymorphism in the cdna producing a heteroduplex running at a size close to the homoduplex product