I think no matter what, you have to make a decision on how you are going to add or integrate the two criteria.  I mean, there are a lot of ways you could do, and you have to decide which you like best..

A simple approach would be to rank-transform the scores within each criteria.  This way, the highest score of criteria A is 1, the next is 2, and so on.  With this approach the only thing that will matter is the ranking.  One issue:  If the scores were 100, 5,4,3,2...  Then this counts as Rank 1,2,3,4,5.  It ignores that the first score was so much higher than the others.

You could also just linearly transform each scale.  That is, make the highest in Score A = 100 and the lowest = 0, and then calculate the rest of the values relative to the original maximum and minimum.  This will allow a score of 100 on Score A to swamp a Score of 40 on A and a score of 40 on B.

A very simple method is using an arithmetical or geometrical weighted mean. This way you give a value to every item in a same scale (Ej. from 0 to 5) according to each criteria. To rank, you obtain a global weighted mean according to the level of importance you give to every criteria. For instance: Sum(Xi*Wj), where Xi are the value of each item "i"according to a criteria "j", and Wj is the level of importance of each criteria "j" under the restriction: (Sum(Wj)=1). If you use different scales, then you have to first normalize the variables into a compatible scale. To do that you can use this formula: Ni=(Xi-min Xi)/(Max Xi- Min Xi), where Ni is the normalized value of each item, min and max Xi are the minimun and maximum values in the original scale of each item Xi. This normalized value will be in the range (0;1)  and you don´t have to worry about an out-of-bound error.

Then you obtain a mean value for each item that allows you to establish a rank. If you want to, you can establish a proceeding to obtain the levels of importance or weight, by using a more appropriate method, (Ej.: Saaty-Dihn). You can also use many OWA operators.

Thanks Eddy and Salvatore for your responses. I have figured out that weighted product method is working quite well (after normalizing data).

@Salvatore, could you please elaborate on how to make best combinations, after rank transform ? What if there is a tie in ranks?

Yes, if you use ranked data, the chance of getting a tie is greater, and you'd have to have some provision for dealing with ties.  Report them as tied?  Or have a tie-breaking procedure that looks at the original scores?

Could you please elaborate on how to find the combination on the basis of these ranks? I have 40 technologies which I want to rank on the basis of best combination between two indicators. I also recently figured out that normalization is not a good starting approach as it does not account for negative valued indicators. For example, if there is a rating of -5 to 5 for every indicator and some units have their value as negative, then normalization just simply assigns a value from 0 to 1 on the basis of max and min. I am not getting good results with this approach.

What I was thinking for a ranking algorithm was just very simple: rank one indicator and rank the other, add them, and have a method to deal with ties.

For a general normalization approach, you could transform everything to a -5 to 5 scale or a -1 to 1 scale.  It shouldn't matter if a scale goes from -5 to 5 or from 0 to 10 or from 0 to 1, but one may be psychology better to present your data.

I'm not sure what it is in your data that making this a difficult question.  From your description, it sounds like an trivial problem.  You have two scales to add together.  If they are on different scales, you need to make them commensurate in some way. ...?

My data is rating data : 3 indicators for 40 technologies , rated between -5 to 5. So in that case they are comparable . The query is that I have to rank technologies as per different combinations , like high on indicator A and B ( C not relevant here ) , high on B C indicators and similarly for A and C indicators . If I am normalising this data , then this negative is being converted to a value between 0 to 1 . For me negative is bad , higher positive values are preferred. In case of combinations , suppose where one indicator is positive and other negative , since negative becomes more than 0 , this item is getting a high rank . Hence that " negative " score , which is considered detrimental, is not being accounted for . It's also a major problem in case of additive based measures.

What if I simply take a product of the two indicators , without normalising , and rank incorporating a tie breaker ? I am getting sound results with this. 

For unique ranking, I came across this simple excel command : Rank(number, ref, order)+ countif(range, criteria)-1 

Though it helps to form unique ranking, I don't find any sound logic in this. I mean if two technologies have equal scores, they stand at same level. Unique ranking is just mathematical manipulation. What is your feedback on this ?

I don't entirely understand the excel formula, so I won't comment specifically.  But I agree with your general thought:  if two technologies have the same score, they should get the same rank.

Since the scales are comparable, I think the best approach is to keep things simple.

So, if a Technology has a score of 5 and of - 3, it isn't okay that its final score is 2 ?