I have used the MDR method (www.multifactordimensionalityreduction.org/) to characterize gene-gene interactions in our articles (Please check their Methods for further details):
Its hard to comment on which is the best method but I can share my approach which I follow in pursuit of understanding gene-gene interactions.
The strategy used, depends on the sample size which is first to get affected during a stratified analysis and so the outcome of these analysis should be validated using other techniques as well.
I believe your trying to associate few SNPs with disease risk and also wish to identify plausible gene-gene interactions. In that case, you must have entered your SNP data as a categorical variable into your database (say created using SPSS). Now if you wish to identify interactions, you could convert your categorical data into numbers.
On how to convert categorical data into numerical, I would suggest the link below
Initially doing such a thing may look like a mammoth task (if your sample size is huge and/or if you have many SNPs) but once done - this data can be statistically analyzed in a number of ways to make sensible conclusions (This is my 1st approach).
But what if you dnt want to get into this trouble or want to get a preliminary idea ?
In that case, as suggested by Marcelo - you can use computational predictions (This is the second thing I do)
Under this header you can do either of the two things
1. MDR (Multifactorial dimension reduction) test - Very easy to do
If you have never done this before and want to start anew, please visit
http://www.epistasisblog.org/
This is a friendly site with all necessary information and download links. Dnt worry - there is no need to create a new database. The software will ask for a 'tab-delimited' text file (having all data) which can be easily created from your SPSS/ xlsx/xls file.
2. I will also suggest using CART - classification and regression tree (There are may tutorial videos on YouTube)
To do this you can try 'Salford Predictive Modeler' which is a great software (requires a subscription) but I believe there will be free alternatives as well.
Why use this - It will stratifies data according to the most important classifiers, so when you study the classification tree, you will get an idea which SNPs are interacting.
3. Finally you can also use receiver operating characteristic ((ROC) curve to check for a statistical increase in the area under curve (AUC) to predict interaction. For instance the AUC will increase if two SNPs interact and elevate the risk of a disease as against their individual AUCs. There are many software platforms but I prefer SPSS and GraphPad Prism.
To check for statistical difference between ROCs, you can use this calculator
http://vassarstats.net/roc_comp.html
You may be also interested in checking linkage (Its wise to check for linkage when you have multiple SNPs).
I will like to conclude by saying that this is my approach of investigating gene-gene & gene-environment interactions. There may be even simpler tricks, so keep looking.