The question whether technical analysis [TA] [or 'charting' as referred to by the less initiated] works has been asked many, many times. Countless studies have been done, some of which show that TA does make a significant difference, and some show that TA does not make any difference to investment performance. For more details of such studies, see: http://en.wikipedia.org/wiki/Technical_analysis . I will use this short article just to recollect my personal experience with TA.
For more than a decade, I studied TA, and tried to apply it. I studied all manner of TA from classical charting, to Candlesticks, to the hundreds of Indicators from Moving Averages, to RSI, MACD and the more esoteric ones ones that are constantly being invented. I read about Elliot Waves and Gann Charts, and how some even used astrology to analyse the markets. I delved into Neural Networks, Genetic Algorithms and all manner of pattern recognition, classification and forecasting techniques to feed the Price [High,Low, Open, Close] and Volume data into to see if I could derive anything useful.
After all this, I realized that the investment world is much more complex than can be modeled by just Price and Volume Data. Investment decisions and prices of stocks are still governed by (1) Liquidity and the flow of Funds (2) Fundamentals like profits, dividends and the macroeconomic interest rate (3) investor behavior. The Technical Analyst says that all these are reflected in the charts, but the charts are not predictive. They can only show what has happened or what is happening. Below are the reasons why I think TA by itself is not complete:
1. TA and charting should basically be viewed only as a visualization aid, for that is what it really is. The charts and indicators compress, smooth, filter, normalize etc the data to help us visualize the market using colors, shape and metrics like % change to aid us in coming to some 'general' conclusion about the situation in the market. The values of a TA Indicator, normalized to %, or - 1 to +1 makes it easier for us to grasp the situation than figures put in a Table. The slope of an indicator enables us to visualize the rate of change, and an oscillator with base at zero makes it easier for us to visualize cyclical movements. Candle Sticks as invented by commodity traders in 18th century Japan are a stroke of genius for capturing and encapsulating market information time series.
2. The hundreds of Indicators basically fall into trend indicators and oscillator indicators, to describe the markets. There is not much difference between many of these indicators. And every new Indicator that is 'invented' is merely a variation of the original group that includes: Moving Averages, MACD, RSI, DMI, Stochastics, OBV and Momentum. TA says, follow the trend but in all cases the user has to define what constitutes a trend in his context.
3. If information is not there, it cannot be squeezed out from nothing and if information is there, processing it will make you lose some. AlsoTechnical Analysis depends on the length of the time period scale that you want to analyze. Moving Averages invariably have lag despite products on the market that claim to eliminate lag. Any attempts to put a predictive element into TA is inherently doomed to failure, and normal TA does not usually claim to be predictive. But exponents of techniques such as Elliot Waves and Gann charts do claim to be predictive, as do those who look for head-and-shoulders in charts or Tombstones and Falling Swords in Candle Stick analysis.
4. There really is no pattern in the markets much of the time. Neural-Networks I ran confirmed this. Where there is a pattern it exists for a fleeting moment and may not repeat. I do not believe in cycles that are regular and sufficiently distinct to be predictive. From my experience using digital signal processing, compression and de-noising techniques [such as Wavelets] to detect cycles, I find cycles appearing and disappearing, morphing into other cycles, and cycles nested within cycles.
5. The character of the market changes constantly. This change is due to fundamental changes in the macroeconomic environment. In 2000, events which led to the dot com bust derailed all market predictions and radically changed the character of the market. Right now, happenings in the Debt market, the glut of cash awash in the world seeking investment returns, the U.S. dollar weakness, the abundance of derivatives of all manner and the action of private equity acquiring debt to acquire Companies. All this will cause great changes in the market and may precipitate a a financial avalanche though nobody knows when. One success due to use of a certain TA method may be useless the next time.
6. One thing that we can be sure of: The stock market is a Complex Adaptive Sytem [CAP]with all the characteristics of such systems. See Wikipedia: http://en.wikipedia.org/wiki/Complex_Adaptive_System . TA cannot adequately model a CAP. Studying the characteristics of a CAP is useful. In Nature, many things are CAP. The stock market is like a super-organism, made up of millions of individuals each doing their own thing, but with the sole aim of making a profit. The characteristics of CAP like having feedback loops, fractals and self-similarity, self-organization, distribution and decentralization and being non-linear, adaptive and co-evolutionary apply to the stock market.
7. With advent of powerful computers, software and good data, it is possible and much better to model the markets as a CAP using fundamentals, but including in the model*, technical measurements such as Momentum [rate of change], Beta and Liquidity. Quantitative Models based on statistical probability make it possible to forecast, classify and do simulation. Such a type of model can accomodate change in market character by a regular re-estimation of model parameters. Take a look at www.valuengine.com
* Advanced models can use more robust and generalized regression algorithms.  non-Gaussian probability distributions  Fuzzy-Logic or some form of non-Euclidean distance methods for clustering  Advanced data-processing methods for de-noising and compression such as Wavelets.  newer algorithms for feature isolation and extraction based on Principal Component Analysis but more akin to the Self-Organizing Map [SOM] of Kohonen.  the latest techniques borrowed from Meteorology [weather forecasting] for update of forecast. Genetic Algorithms for optimization processes where you want to avoid Local Minima  Full-Scale simulation for those aspects which cannot be modeled in a closed-form.
Warren Buffet once said about TA: "I realized technical analysis didn't work when I turned the charts upside down and didn't get a different answer" and "If past history was all there was to the game, the richest people would be librarians."
TA is not completely useless. The shorter the period under analysis, and the less the market is involved with fundamentals, the more useful TA becomes. In FOREX and Commodities, and day-trading for example, knowledge of TA is a pre-requisite. Indicators that measure momentum,beta [the sensitivity of a stock to the movements of the general market as represented by an Index], volume of trades, ratio of declines/advances, intensity of trend, quality of price changes etc are useful. Trend may not be predictive, but the general rule in Nature- that if something is trending, it will continue to trend [due to the phenomenon of positive feedback reinforcement] unless something happens to slow down or reverse the trend applies. Supports and Resistances, though not something to bet your house on, are the manifestation of the natural tendency of data to cluster around certain points because of on human psychological behavior viz such as round numbers, significant numbers, certain days of the week, month etc.
To conclude: View TA not as some mumbo-jumbo that can predict the market, but a form of information compression and visualization aid, which like all of Mankind's attempts to manage information overload should be lauded.