diff --git a/README.md b/README.md
index 8da07a4..31887d1 100644
--- a/README.md
+++ b/README.md
@@ -7,5 +7,9 @@
 This is a list of the rules and formatting transformations performed by Runic:
 
  - No trailing whitespace
- - Normalized line endings (`\r\n` -> `\n`) (TODO: Is this bad on Windows with Git's autocrlf?)
- - Hex/octal/binary literals are padded with zeroes
+ - Normalized line endings (`\r\n` -> `\n`) (TODO: Is this bad on Windows with Git's autocrlf? gofmt does it...)
+ - Hex/octal/binary literals are padded with zeroes to better highlight the resulting UInt
+   type
+ - Floating point literals are normalized to always have an integral and fractional part.
+   `E`-exponents are normalized to `e`-exponents. Unnecessary trailing/leading zeros from
+   integral, fractional, and exponent parts are removed.
diff --git a/src/Runic.jl b/src/Runic.jl
index c9c0901..f6d4732 100644
--- a/src/Runic.jl
+++ b/src/Runic.jl
@@ -190,6 +190,7 @@ function format_node!(ctx::Context, node::JuliaSyntax.GreenNode)::Union{JuliaSyn
     @return_something trim_trailing_whitespace(ctx, node)
     @return_something format_hex_literals(ctx, node)
     @return_something format_oct_literals(ctx, node)
+    @return_something format_float_literals(ctx, node)
 
     # If the node is unchanged at this point, just keep going.
 
diff --git a/src/runestone.jl b/src/runestone.jl
index 85cfd62..d20ce62 100644
--- a/src/runestone.jl
+++ b/src/runestone.jl
@@ -87,3 +87,55 @@ function format_oct_literals(ctx::Context, node::JuliaSyntax.GreenNode)
     node′ = JuliaSyntax.GreenNode(JuliaSyntax.head(node), nb, ())
     return node′
 end
+
+function format_float_literals(ctx::Context, node::JuliaSyntax.GreenNode)
+    JuliaSyntax.kind(node) in KSet"Float Float32" || return nothing
+    @assert JuliaSyntax.flags(node) == 0
+    @assert !JuliaSyntax.haschildren(node)
+    str = String(node_bytes(ctx, node))
+    # Check and shortcut the happy path first
+    r = r"""
+    ^
+    (?:(?:[1-9]\d*)|0)   # Non-zero followed by any digit, or just a single zero
+    \.                   # Decimal point
+    (?:(?:\d*[1-9])|0)   # Any digit with a final nonzero, or just a single zero
+    (?:[ef][+-]?[1-9]\d*)?
+    $
+    """x
+    if occursin(r, str)
+        return nothing
+    end
+    if occursin('_', str) || startswith(str, "0x")
+        # TODO: Hex floats and floats with underscores are ignored
+        return nothing
+    end
+    # Split up the pieces
+    r = r"^(?<int>\d*)(?:\.?(?<frac>\d*))?(?:(?<epm>[eEf][+-]?)(?<exp>\d+))?$"
+    m = match(r, str)
+    io = IOBuffer() # TODO: Could be reused?
+    # Strip leading zeros from integral part
+    int_part = isempty(m[:int]) ? "0" : m[:int]
+    int_part = replace(int_part, r"^0*((?:[1-9]\d*)|0)$" => s"\1")
+    write(io, int_part)
+    # Always write the decimal point
+    write(io, ".")
+    # Strip trailing zeros from fractional part
+    frac_part = isempty(m[:frac]) ? "0" : m[:frac]
+    frac_part = replace(frac_part, r"^((?:\d*[1-9])|0)0*$" => s"\1")
+    write(io, frac_part)
+    # Write the exponent part
+    if m[:epm] !== nothing
+        write(io, replace(m[:epm], "E" => "e"))
+        @assert m[:exp] !== nothing
+        # Strip leading zeros from integral part
+        exp_part = isempty(m[:exp]) ? "0" : m[:exp]
+        exp_part = replace(exp_part, r"^0*((?:[1-9]\d*)|0)$" => s"\1")
+        write(io, exp_part)
+    end
+    bytes = take!(io)
+    nb = write_and_reset(ctx, bytes)
+    @assert nb == length(bytes)
+    # Create new node and return it
+    node′ = JuliaSyntax.GreenNode(JuliaSyntax.head(node), nb, ())
+    return node′
+end
diff --git a/test/runtests.jl b/test/runtests.jl
index 73f2538..3bfd82a 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -53,22 +53,39 @@ end
         "0o4" * z(42) => "0o4" * z(42), # typemax(UInt128) + 1
         "0o7" * z(42) => "0o7" * z(42),
     ]
-    # Test the test cases :)
     mod = Module()
     for (a, b) in test_cases
         c = Core.eval(mod, Meta.parse(a))
         d = Core.eval(mod, Meta.parse(b))
         @test c == d
         @test typeof(c) == typeof(d)
+        @test format_string(a) == b
     end
-    # Join all cases to a single string so that we only need to call the formatter once
-    input_str = let io = IOBuffer()
-        join(io, (case.first for case in test_cases), '\n')
-        String(take!(io))
-    end
-    output_str = let io = IOBuffer()
-        join(io, (case.second for case in test_cases), '\n')
-        String(take!(io))
+end
+
+@testset "Floating point literals" begin
+    test_cases = [
+        ["1.0", "1.", "01.", "001.", "001.00", "1.00"] => "1.0",
+        ["0.1", ".1", ".10", ".100", "00.100", "0.10"] => "0.1",
+        ["1.1", "01.1", "1.10", "1.100", "001.100", "01.10"] => "1.1",
+        ["1e3", "01e3", "01.e3", "1.e3", "1.000e3", "01.00e3"] => "1.0e3",
+        ["1e+3", "01e+3", "01.e+3", "1.e+3", "1.000e+3", "01.00e+3"] => "1.0e+3",
+        ["1e-3", "01e-3", "01.e-3", "1.e-3", "1.000e-3", "01.00e-3"] => "1.0e-3",
+        ["1E3", "01E3", "01.E3", "1.E3", "1.000E3", "01.00E3"] => "1.0e3",
+        ["1E+3", "01E+3", "01.E+3", "1.E+3", "1.000E+3", "01.00E+3"] => "1.0e+3",
+        ["1E-3", "01E-3", "01.E-3", "1.E-3", "1.000E-3", "01.00E-3"] => "1.0e-3",
+        ["1f3", "01f3", "01.f3", "1.f3", "1.000f3", "01.00f3"] => "1.0f3",
+        ["1f+3", "01f+3", "01.f+3", "1.f+3", "1.000f+3", "01.00f+3"] => "1.0f+3",
+        ["1f-3", "01f-3", "01.f-3", "1.f-3", "1.000f-3", "01.00f-3"] => "1.0f-3",
+    ]
+    mod = Module()
+    for (as, b) in test_cases
+        for a in as
+            c = Core.eval(mod, Meta.parse(a))
+            d = Core.eval(mod, Meta.parse(b))
+            @test c == d
+            @test typeof(c) == typeof(d)
+            @test format_string(a) == b
+        end
     end
-    @test format_string(input_str) == output_str
 end